Awesome AI for CAD

A curated catalog of papers, datasets, and resources on AI for Computer-Aided Design.

---

Taxonomy

AI for CAD
├── 2D CAD Intelligence
│   ├── Symbol Detection & Spotting
│   ├── Text & Annotation Extraction
│   ├── Drawing Understanding & Benchmarks
│   ├── Compliance Checking
│   ├── Floor Plan Generation & Analysis
│   ├── Vectorization & Digitization
│   ├── P&ID Diagram Intelligence
│   └── Electrical & Circuit Schematics
├── 3D CAD Generation
│   ├── Autoregressive Sequence Models
│   ├── Diffusion-Based Generation
│   ├── LLM/VLM-Based Generation
│   ├── B-Rep & CSG Generation
│   ├── Point Cloud / Image / Sketch to CAD
│   ├── CAD Editing & Assembly
│   └── Shape Programs & Procedural Generation
├── CAD Understanding
│   ├── B-Rep Representation Learning
│   ├── Multi-Modal CAD Representations
│   ├── Machining Feature Recognition
│   ├── Retrieval & Classification
│   └── Assembly Understanding
├── Simulation & Optimization
│   ├── Neural Operators & FEA Surrogates
│   ├── Computational Fluid Dynamics
│   ├── Physics-Informed Neural Networks
│   ├── Topology Optimization
│   └── AI-Driven Generative Design
└── Manufacturing-Aware Design
    ├── Design for Manufacturing
    ├── Design for Additive Manufacturing
    ├── Assembly Planning & Tolerance
    └── CAD/CAM Integration

Contents

• Surveys
• Foundational CAD AI Papers
• CAD Representations and Foundations
• 2D CAD and Drawing Intelligence
• 3D CAD Generation and Reconstruction
• CAD Understanding and Retrieval
• Simulation and Design Optimization
• Manufacturing-Aware Design
• Challenges and Future Directions
• Datasets and Benchmarks
• Tools and Platforms
• Contributing
• Catalog Metadata

---

Surveys

Overview and survey papers covering AI for CAD and related 3D generation domains.

• 3D Shape Generation: A Survey — Surveys methods and benchmarks for 3D shape generation across representations. Zibo Zhao et al., arXiv 2025. [2506.22678]
• A Survey on Deep Learning in 3D CAD Reconstruction — Reviews deep learning approaches for reconstructing 3D CAD models from various inputs. Ding et al., Applied Sciences (MDPI) 2025. [Paper]
• Advances in 3D Generation: A Survey — Comprehensive survey of recent advances in 3D content generation techniques. Xiaoyu Li et al., arXiv 2024. [2401.17807]
• Diffusion Models in 3D Vision: A Survey — Surveys applications of diffusion models to 3D vision tasks. Zhen Wang et al., arXiv 2024. [2410.04738]
• A Survey On Text-to-3D Contents Generation In The Wild — Reviews text-to-3D generation methods for open-domain content creation. Chenhan Jiang et al., arXiv 2024. [2405.09431]
• Applications of Artificial Intelligence in the AEC Industry: A Review — Reviews AI applications in architecture, engineering, and construction workflows. Zheng et al., Journal of Asian Architecture and Building Engineering 2024. [Paper]
• LLM4CAD: Multi-Modal Large Language Models for 3D Computer-Aided Design Generation — Proposes using multi-modal LLMs for generating 3D CAD models. Various, ASME IDETC-CIE 2024. [Paper]

---

Foundational CAD AI Papers

Curated CAD-specific anchors that introduced reusable datasets, representations, or modeling paradigms later work repeatedly builds on. This section is intentionally selective; the longer category lists below remain chronological.

Related general AI background references

• Multi-Scale Latent Diffusion with Mamba+ for Complex Parametric Sequence — Generates complex parametric CAD sequences using multi-scale latent diffusion combined with Mamba+ architecture. Liyuan Deng, Yunpeng Bai, Yongkang Dai et al., arXiv 2025. [2511.17647]
• CAD-Coder: Text-to-CAD Generation with Chain-of-Thought and Geometric Reward — Converts text descriptions to CAD models using chain-of-thought reasoning and geometric reward signals. Yandong Guan, Xilin Wang, Ximing Xing et al., arXiv 2025. [2505.19713]
• An Open-Source Vision-Language Model for Computer-Aided Design Code Generation — Proposes an open-source vision-language model that generates CAD code from visual inputs. Anna C. Doris, Md Ferdous Alam, Amin Heyrani Nobari et al., arXiv 2025. [2505.14646]
• OpenECAD: An Efficient Visual Language Model for Editable 3D-CAD Design — Presents an efficient visual language model for generating editable 3D CAD designs. Zhe Yuan, Jianqi Shi, Yanhong Huang, arXiv 2024. [2406.09913]
• Geometric Deep Learning: Grids, Groups, Graphs, Geodesics, and Gauges — Unifies geometric deep learning approaches through symmetry and invariance principles across domains. Bronstein et al., arXiv / IEEE Signal Processing Magazine 2021. [2104.13478]
• Dynamic Graph CNN for Learning on Point Clouds — Introduces EdgeConv module that dynamically computes graphs for learning on point clouds. Wang et al., ACM TOG 2019. [1801.07829]
• PointNet: Deep Learning on Point Sets for 3D Classification and Segmentation — Proposes a unified architecture for directly consuming unordered point sets for 3D tasks. Qi et al., CVPR 2017. [1612.00593]
• PointNet++: Deep Hierarchical Feature Learning on Point Sets in a Metric Space — Extends PointNet with hierarchical feature learning to capture local geometric structures. Qi et al., NeurIPS 2017. [1706.02413]

---

CAD Representations and Foundations

Papers establishing core CAD representation paradigms (B-rep, CSG, sequence, code) that form the basis for generative and understanding methods.

Representative anchors: CSGNet for neural CSG parsing; BRepNet for topological message passing; SkexGen for disentangled CAD codebooks.

• CADIR: A Cross-Backend Editable Intermediate Representation for Agentic CAD Generation — Represents CAD construction as an executable graph with explicit dependencies, diagnostics, and cross-backend feature reconstruction. Yu Liu, Jingzhe Ni, Yiming Chen et al., arXiv 2026. [2608.00891]
• DualBrep: A Dual-Field Continuous Representation for B-rep Modelling — Encodes B-rep geometry and topology jointly in a continuous dual-field representation. Yilin Liu, Pradeep Jayaraman, Chinthala Reddy et al., arXiv 2026. [2606.31579]
• Bridging CAD and Data-Driven Design: Attributed Feature Graphs for Engineering Design — Preserves parametric features and dependencies for interpretable CAD-native surrogate modeling. Abhishek Indupally, Ibraheem Alawadhi, Satchit Ramnath et al., ASME IDETC-CIE 2026. [2606.06405]
• Masked BRep Autoencoder via Hierarchical Graph Transformer — Learns B-rep representations through masked autoencoding on hierarchical graph transformers. Xu et al., arXiv 2026. [2603.14927]
• CAD-Coder: A New Paradigm for CAD Generation with Scalable Large Model Capabilities — Generates CAD models by leveraging scalable large language model code generation. Li et al., arXiv 2025. [2505.06507]
• A Language Model-Driven Multi-Agent System for Collaborative Design — Proposes a multi-agent system powered by language models for collaborative CAD design. Makatura et al., arXiv 2025. [2503.04417]
• Generating Constructive Solid Geometry Instead of Meshes by Fine-Tuning a Code-Generation LLM — Fine-tunes a code-generation LLM to output CSG representations instead of meshes. Skalic et al., arXiv 2024. [2411.15279]
• HNC-CAD: Hierarchical Neural Coding for Controllable CAD Model Generation — Introduces hierarchical neural coding for controllable generation of CAD models. Xu et al., CVPR 2023. [2307.00149]
• D2CSG: Unsupervised Learning of Compact CSG Trees with Dual Complements and Dropouts — Learns compact CSG tree representations from point clouds using dual complement operations and dropout regularization. Kania et al., NeurIPS 2023. [2301.11497]
• Self-Supervised CAD Reconstruction by Learning Sketch-Extrude Operations — Reconstructs CAD models from point clouds by learning sketch-and-extrude modeling sequences without supervision. Ren et al., arXiv 2023. [2303.10613]
• SkexGen: Autoregressive Generation of CAD Construction Sequences with Disentangled Codebooks — Autoregressively generates CAD sketch-extrude sequences using disentangled codebooks for topology and geometry. Xu et al., ICML 2022. [2207.04632]
• BRepNet: A Topological Message Passing System for Solid Models — Proposes message passing on B-Rep topology graphs for learning directly from CAD solid models. Lambourne et al., CVPR 2021 (Oral). [2104.00706]
• UV-Net: Learning from Boundary Representations — Learns from B-Rep UV-domain surface parameterizations for CAD model understanding tasks. Jayaraman et al., CVPR 2021. [2006.10211]
• UCSG-Net: Unsupervised Discovering of Constructive Solid Geometry Tree — Discovers CSG tree structures from shapes in an unsupervised manner without ground-truth programs. Kania et al., NeurIPS 2020. [2006.09102]
• CSGNet: Neural Shape Parser for Constructive Solid Geometry — Parses 2D/3D shapes into CSG primitive programs using a neural network shape parser. Sharma et al., CVPR 2018. [1712.08290]
• B-Rep Distance Functions (BR-DF): How to Represent a B-Rep Model by Volumetric Distance Functions? — Represents B-rep CAD models with volumetric distance functions as a new geometric representation. Zhang et al., arXiv 2025. [2511.14870]

---

2D CAD and Drawing Intelligence

AI methods for interpreting, analyzing, and generating 2D engineering drawings, floor plans, P&ID diagrams, and circuit schematics.

Representative anchors: SymPoint for panoptic symbol spotting; FloorPlanCAD for drawing datasets; GAT-CADNet for graph-based symbol detection.

Symbol Detection and Spotting

• Text-Aided Multi-Modal Panoptic Symbol Spotting for CAD Floor Plan Drawings — Fuses vector primitives and layered text for panoptic symbol spotting in CAD drawings. Yan Gong, Bohao Li, Bowen Du et al., arXiv 2026. [2607.12678]
• Point or Line? — Proposes line-based representation as an alternative to point-based methods for panoptic symbol spotting. Xingguang Wei, Haomin Wang, Shenglong Ye et al., arXiv 2025. [2505.23395]
• Text-Enhanced Panoptic Symbol Spotting — Integrates textual information to improve panoptic symbol spotting performance in CAD drawings. Xianlin Liu, Yan Gong, Bohao Li et al., BESC 2025. [2510.11091]
• Relative Drawing Identification Complexity — Shows that drawing identification difficulty remains consistent across modalities in vision-language models. Authors, arXiv 2025. [2505.10583]
• Architectural Practice Process and Artificial Intelligence — Examines how AI reshapes evolving architectural design practice processes. Authors, arXiv 2025. [2507.23653]
• Symbol as Points — Represents symbols as points for unified panoptic symbol spotting via point-based detection. Wenlong Liu, Tianyu Yang, Yuhan Wang et al., ICLR 2024. [2401.10556]
• SymPoint Revolutionized — Boosts panoptic symbol spotting accuracy by enhancing layer-wise feature representations in SymPoint. Wenlong Liu, Tianyu Yang, Qizhi Yu et al., arXiv 2024. [2407.01928]
• CADSpotting — Presents a robust panoptic symbol spotting method designed to scale to large-scale CAD drawings. Fuyi Yang, Jiazuo Mu, Yanshun Zhang et al., arXiv 2024. [2412.07377]
• Generative AI in the Construction Industry: A State-of-the-art Analysis — Surveys generative AI applications across construction tasks including design, planning, and documentation. Ridwan Taiwo, Idris Temitope Bello, Sulemana Fatoama Abdulai et al., arXiv 2024. [2402.09939]
• Exploring Gen-AI applications in building research and industry: A review — Reviews generative AI use cases in building design, energy modeling, and facility management. Hanlong Wan, Jian Zhang, Yan Chen et al., arXiv 2024. [2410.01098]
• GAT-CADNet: Graph Attention Network for Panoptic Symbol Spotting in CAD Drawings — Proposes a graph attention network for panoptic symbol spotting in CAD drawings. Zhaohua Zheng, Jianfang Li, Lingjie Zhu et al., arXiv 2022. [2201.00625]
• Automatic Detection and Classification of Symbols in Engineering Drawings — Detects and classifies symbols in engineering drawings using deep learning methods. Sourish Sarkar, Pranav Pandey, Sibsambhu Kar, arXiv 2022. [2204.13277]
• Discovering Design Concepts for CAD Sketches — Learns reusable design concepts from CAD sketch datasets via unsupervised discovery. Yuezhi Yang, Hao Pan, NeurIPS 2022. [2210.14451]
• The Scope for AI-Augmented Interpretation of Building Blueprints in Commercial and Industrial Property Insurance — Explores AI-driven blueprint interpretation to automate property insurance risk assessment. Long Chen, Mao Ye, Alistair Milne et al., arXiv 2022. [2205.01671]
• An Automated Engineering Assistant — Presents an automated system for recognizing and interpreting symbols in technical drawings. Dries Van Daele, Nicholas Decleyre, Herman Dubois et al., arXiv 2019. [1909.08552]

Text and Annotation Extraction

• Context-Aware Mapping of 2D Drawing Annotations to 3D CAD Features Using LLM-Assisted Reasoning for Manufacturing Automation — Maps 2D drawing annotations to 3D CAD features using LLM-assisted reasoning for manufacturing workflows. Muhammad Tayyab Khan, Lequn Chen, Wenhe Feng et al., arXiv 2026. [2602.18296]
• A Multi-Stage Hybrid Framework for Automated Interpretation of Multi-View Engineering Drawings Using Vision Language Model — Proposes a multi-stage hybrid framework combining vision-language models to interpret multi-view engineering drawings. Muhammad Tayyab Khan, Zane Yong, Lequn Chen et al., ICIEA 2026. [2510.21862]
• Automated Parsing of Engineering Drawings for Structured Information Extraction Using a Fine-tuned Document Understanding Transformer — Fine-tunes a document understanding transformer to parse engineering drawings into structured information. Muhammad Tayyab Khan, Zane Yong, Lequn Chen et al., IEEE IEEM 2025. [2505.01530]
• A Hybrid Vision-Language Framework for Parsing 2D Engineering Drawings into Structured Manufacturing Knowledge — Parses 2D engineering drawings into structured manufacturing knowledge using a hybrid vision-language framework. Authors, arXiv 2025. [2506.17374]
• Fine-Tuning Vision-Language Model for Automated Engineering Drawing Information Extraction — Fine-tunes a vision-language model to automate information extraction from engineering drawings. Muhammad Tayyab Khan, Lequn Chen, Ye Han Ng et al., ICIAI 2025. [2411.03707]

Drawing Understanding and Benchmarks

• CrossProjection: Geometric Grounding Beyond Viewpoint Change in Architectural Drawings — Audits whether vision-language models preserve component identity and explicitly localize geometry across plans, sections, and elevations. Kaho Li, Pengyu Zeng, Yuqin Dai et al., arXiv 2026. [2608.00473]
• Benchmarking Deep Learning Approaches for AEC Engineering Drawing Layout Detection and Information Extraction — Benchmarks layout detection and information extraction on structured AEC engineering drawings. Tianyang Huang, Alessio Lombardi, Ahmed Elnagar et al., EC3 2026. [2607.18997]
• MechVQA: Benchmarking and Enhancing Multimodal LLMs on Comprehensive Mechanical Drawing Understanding — Benchmarks recognition, reasoning, and judgment on 3.3K mechanical drawings and 21K question-answer pairs. Qian Kou, Xiaofeng Shi, Yulin Li et al., ICML 2026. [2605.30794]
• AEC-Bench — Evaluates agentic AI systems on multimodal tasks in architecture, engineering, and construction. Harsh Mankodiya, Chase Gallik, Theodoros Galanos et al., arXiv 2026. [2603.29199]
• Blueprint — Multimodal retrieval system for complex engineering drawings and technical documents. Authors, arXiv 2026. [2602.13345]
• Advancing Multimodal LLM Evaluation of Engineering Documentation — Enhances retrieval-augmented evaluation of multimodal LLMs on engineering documents. Authors, arXiv 2026. [2604.09552]
• Solving Idealized Beam Models from Hand-Drawn Drawings — Extracts structural beam models automatically from hand-drawn engineering sketches. Authors, arXiv 2026. [2603.21432]
• Blueprint-Bench — Compares spatial intelligence of LLMs, agents, and image models on blueprint tasks. Lukas Petersson, Axel Backlund, Axel Wennstöm et al., Submitted to ICLR 2026. [2509.25229]
• AECBench — Hierarchical benchmark for evaluating LLM knowledge in architecture, engineering, and construction. Chen Liang, Zhaoqi Huang, Haofen Wang et al., Advanced Engineering Informatics 2025. [2509.18776]
• VectorGraphNET — Uses graph attention networks for accurate segmentation of complex technical drawings. Andrea Carrara, Stavros Nousias, André Borrmann, arXiv 2024. [2410.01336]
• DesignQA: A Multimodal Benchmark for Evaluating Large Language Models' Understanding of Engineering Documentation — Introduces a multimodal benchmark to evaluate LLMs on engineering drawing comprehension tasks. Anna C. Doris, Daniele Grandi, Ryan Tomich et al., ASME JCISE 2024. [2404.07917]
• Prediction of Visual Relations in Engineering Drawings — Proposes methods to predict spatial and semantic relations between entities in engineering drawings. Chao Gu, Ke Lin, Yiyang Luo et al., arXiv 2024. [2409.00909]

2D-3D Annotation Mapping

• Drawing-Recode: Annotation Grounding for Parametric CAD Code Generation from Raster 2D CAD Drawings — Grounds dimensional annotations in raster engineering drawings to generate structured parametric CAD code. Mingi Kim, Yongjun Kim, Hyungki Kim, arXiv 2026. [2607.27558]
• CAD2Program: From 2D CAD Drawings to 3D Parametric Models — Converts 2D CAD drawings into 3D parametric modeling programs via learned mappings. Xilin Wang, Jia Zheng, Yuanchao Hu et al., AAAI 2025. [2412.11892]

Compliance Checking

• Automated Facility Enumeration for Building Compliance Checking using Door Detection and Large Language Models — Combines door detection with LLMs to automate facility enumeration for building code compliance. Licheng Zhang, Bach Le, Naveed Akhtar et al., arXiv 2025. [2509.17283]
• Automatic Building Code Review — Automates building code review processes using AI-driven analysis of architectural plans. Authors, arXiv 2025. [2510.02634]
• DRC-Coder: Automated DRC Checker Code Generation Using LLM Autonomous Agent — Uses an LLM autonomous agent to automatically generate design rule checking code. Chen-Chia Chang, Chia-Tung Ho, Yaguang Li et al., ISPD 2025. [2412.05311]
• ARCEAK: An Automated Rule Checking Framework Enhanced with Architectural Knowledge — Proposes a rule checking framework that integrates architectural knowledge for compliance automation. Junyong Chen, Ling-I Wu, Minyu Chen et al., arXiv 2024. [2501.14735]
• CODE-ACCORD: A Corpus of Building Regulatory Data for Rule Generation towards Automatic Compliance Checking — Introduces a building regulatory corpus for generating machine-readable rules for compliance checking. Hansi Hettiarachchi, Amna Dridi, Mohamed Medhat Gaber et al., Scientific Data 2024. [2403.02231]

Floor Plan and Drawing Generation

• Unified Vector Floorplan Generation via Markup Representation — Generates vector floor plans using a unified markup language representation. Authors, arXiv 2026. [2604.04859]
• A Two-Level Codebook Based Network for End-to-End Vector Floorplan Generation — Uses a two-level codebook to encode and generate vector floor plans end-to-end. Authors, arXiv 2026. [2602.07100]
• HouseMind: Tokenization Allows Multimodal Large Language Models to Understand, Generate and Edit Architectural Floor Plans — Enables multimodal LLMs to understand, generate, and edit floor plans via tokenization. Sizhong Qin, Ramon Elias Weber, Xinzheng Lu, CVPR 2026. [2603.11640]
• Controllable End-to-End Vector Floor Plan Generation — Proposes controllable end-to-end generation of vector floor plans with user constraints. Authors, arXiv 2026. [2602.20377]
• Text-to-Layout: A Generative Workflow for Drafting Architectural Floor Plans Using LLMs — Drafts architectural floor plan layouts from natural language descriptions using LLMs. Jayakrishna Duggempudi, Lu Gao, Ahmed Senouci et al., arXiv 2025. [2509.00543]
• Small Building Model: A Transformer for Layout Synthesis in BIM — Applies a transformer architecture for automated layout synthesis in BIM environments. Authors, arXiv 2025. [2512.04832]
• DiffPlanner: Direct Vector Floor Plan Generation with Diffusion — Generates vector floor plans directly using a diffusion-based generative model. Authors, arXiv 2025. [2508.13738]
• Computer-Aided Layout Generation for Building Design: A Survey — Surveys computational methods for automated building layout generation. Authors, Computational Visual Media 2025. [2504.09694]
• HypergraphFormer: Learning Hypergraphs from LLMs for Editable Floor Plan Generation — Generates editable floor plans by learning hypergraph representations distilled from large language models. Klimenko et al., arXiv 2026. [2605.18932]
• Generative Floor Plan Design with LLMs via Reinforcement Learning with Verifiable Rewards — Controls room dimensions and connectivity in floor plan design using LLMs trained with verifiable-reward RL. Lara et al., arXiv 2026. [2605.14117]
• What a Comfortable World: Ergonomic Principles Guided Apartment Layout Generation — Generates apartment layouts guided by ergonomic principles to avoid inefficiencies learned from real data. Nieciecki et al., arXiv 2026. [2604.08411]
• Space Syntax-guided Post-training for Residential Floor Plan Generation — Post-trains floor plan generators with space-syntax guidance for better spatial configurational logic. Jiang et al., arXiv 2026. [2602.22507]
• GFLAN: Generative Functional Layouts — Generates functional floor plan layouts combining combinatorial search with geometric constraint satisfaction. Abouagour et al., arXiv 2025. [2512.16275]

Vectorization and Digitization

• FloorplanVLM: A Vision-Language Model for Floorplan Vectorization — Uses a vision-language model to vectorize floorplan images into structured representations. Authors, arXiv 2026. [2602.06507]
• Drawing2CAD: Sequence-to-Sequence Learning for CAD Generation from Vectorized Drawings — Proposes sequence-to-sequence learning to generate CAD models from vectorized engineering drawings. Feiwei Qin, Shichao Lu, Junhao Hou et al., ACM MM 2025. [2508.18733]
• Line Drawing Pretraining for Efficient, Transferable, and Human-Aligned Vision — Pretrains vision models on line drawings for efficient and human-aligned visual representations. Authors, arXiv 2025. [2508.06696]
• Enhancing Structured Reasoning for Vector Graphics Generation with Reinforcement Learning — Applies reinforcement learning to improve structured reasoning in vector graphics generation. Authors, CVPR 2026. [2505.24499]
• Rendering-Aware Reinforcement Learning for Vector Graphics Generation — Introduces rendering-aware rewards in reinforcement learning for higher-quality vector graphics output. Authors, arXiv 2025. [2505.20793]
• Vector Graphic Animation via Neural Implicits and Video Diffusion Priors — Animates vector graphics using neural implicit representations and video diffusion priors. Authors, arXiv 2025. [2509.07484]
• Advanced Knowledge Extraction of Physical Design Drawings, Translation and Conversion to CAD Formats using Deep Learning — Extracts knowledge from physical design drawings and converts them to CAD formats via deep learning. Jesher Joshua M, Ragav V, Syed Ibrahim S P, arXiv 2024. [2403.11291]
• Evaluating Large Language Models on Vector Graphics Understanding and Generation — Benchmarks LLM capabilities on understanding and generating vector graphics. Authors, arXiv 2024. [2407.10972]
• Tokenizing Strokes for Vector Graphic Synthesis — Introduces a stroke tokenization method for generating vector graphics via sequence modeling. Zecheng Tang, Chenfei Wu, Zekai Zhang et al., arXiv 2024. [2401.17093]
• A Comprehensive End-to-End Computer Vision Framework for Restoration and Recognition of Low-Quality Engineering Drawings — Proposes an end-to-end framework to restore and recognize degraded engineering drawings. Lvyang Yang, Jiankang Zhang, Huaiqiang Li et al., Engineering Applications of Artificial Intelligence 2023. [2312.13620]
• Component Segmentation of Engineering Drawings Using Graph Convolutional Networks — Applies graph convolutional networks to segment components in engineering drawings. Wentai Zhang, Joe Joseph, Yue Yin et al., Computers in Industry 2022. [2212.00290]
• Deep Vectorization of Technical Drawings — Presents a deep learning approach to convert raster technical drawings into vector representations. Egiazarian et al., ECCV 2020. [2003.05471]

P&ID Diagram Intelligence

• GraphRAG for Engineering Diagrams — Applies graph-based retrieval-augmented generation to query and reason over engineering diagram content. Jan Marius Stürmer, Tobias Koch, arXiv 2026. [2603.22528]
• A Closed-Loop, Physics-Aware Agentic Framework for Auto-Generating Chemical Process and Instrumentation Diagrams — Proposes a physics-aware agentic framework that automatically generates chemical process and instrumentation diagrams. Authors, arXiv 2025. [2505.24584]
• Talking like Piping and Instrumentation Diagrams (P&IDs) — Introduces a natural language interface for interpreting and communicating P&ID diagram content. Achmad Anggawirya Alimin, Dominik P. Goldstein, Lukas Schulze Balhorn et al., Systems and Control Transactions 2025. [2502.18928]
• Visual Language Model as a Judge for Object Detection in Industrial Diagrams — Leverages vision-language models to evaluate object detection quality in industrial diagrams. Sanjukta Ghosh, IEEE ICASSP 2026. [2510.03376]
• Advanced Integration of Discrete Line Segments in Digitized P&ID for Continuous Instrument Connectivity — Integrates discrete line segments in digitized P&IDs to reconstruct continuous instrument connectivity paths. Soumya Swarup Prusty, Astha Agarwal, Srinivasan Iyenger, arXiv 2025. [2505.11976]
• Rule-Based Autocorrection of Piping and Instrumentation Diagrams (P&IDs) on Graphs — Applies rule-based methods on graph representations to automatically correct errors in P&IDs. Authors, arXiv 2025. [2502.18493]
• Accelerating Manufacturing Scale-Up from Material Discovery Using Agentic Web Navigation and Retrieval-Augmented AI for Process Engineering Schematics Design — Combines agentic web navigation with retrieval-augmented AI to accelerate process engineering schematic design. Authors, arXiv 2024. [2412.05937]
• An Agentic Approach to Automatic Creation of P&ID Diagrams from Natural Language Descriptions — Uses LLM agents to automatically generate P&ID diagrams from natural language input. Shreeyash Gowaikar, Srinivasan Iyengar, Sameer Segal et al., AAAI 2025 Workshop AI2ASE. [2412.12898]
• Towards Automatic Generation of Piping and Instrumentation Diagrams (P&IDs) with Artificial Intelligence — Explores AI methods for automatically generating P&ID diagrams from process descriptions. Edwin Hirtreiter, Lukas Schulze Balhorn, Artur M. Schweidtmann, arXiv 2022. [2211.05583]
• Digitize-PID: Automatic Digitization of Piping and Instrumentation Diagrams — Proposes a pipeline for automatically digitizing scanned P&ID sheets into structured formats. Shubham Paliwal, Arushi Jain, Monika Sharma et al., PAKDD 2021. [2109.03794]
• OSSR-PID: One-Shot Symbol Recognition in P&ID Sheets using Path Sampling and GCN — Performs one-shot symbol recognition in P&IDs using path sampling and graph convolutional networks. Shubham Paliwal, Monika Sharma, Lovekesh Vig, IJCNN 2021. [2109.03849]
• SynthPID: P&ID Digitization from Topology-Preserving Synthetic Data — Digitizes piping and instrumentation diagrams into process graphs using topology-preserving synthetic training data. Prasad et al., arXiv 2026. [2604.16513]

Electrical and Circuit Schematics

• Learning to Ground Before Reading: Unified PCB Engineering Drawing Parsing with Compact Vision-Language Models — Parses full-page PCB engineering drawings into localized region classes, bounding boxes, and structured text or table content with a compact VLM. Jinghao Liu, Xingrun Liu, Gengchen Sun et al., arXiv 2026. [2608.29268]
• OmniRouting: A Semantic-Coupled Multimodal Benchmark for Constraint-Aware Spatial Reasoning in PCB Routing — Evaluates multimodal models on PCB routing under geometric, electrical, connectivity, and manufacturability constraints. Taiting Lu, Kaiyuan Lin, Ziwei Dong et al., arXiv 2026. [2608.04434]
• SINA: A Circuit Schematic Image-to-Netlist Generator Using Artificial Intelligence — Converts circuit schematic images to netlists using AI-based recognition and extraction. Saoud Aldowaish, Yashwanth Karumanchi, Kai-Chen Chiang et al., arXiv 2026. [2601.22114]
• OmniSch: A Multimodal PCB Schematic Benchmark For Structured Diagram Visual Reasoning — Introduces a multimodal benchmark for visual reasoning over PCB schematic diagrams. Taiting Lu, Kaiyuan Lin, Yuxin Tian et al., arXiv 2026. [2604.00270]
• CircuitLM: A Multi-Agent LLM-Aided Design Framework for Generating Circuit Schematics from Natural Language Prompts — Generates circuit schematics from natural language using a multi-agent LLM framework. Authors, arXiv 2026. [2601.04505]
• PCBSchemaGen: Constraint-Guided Schematic Design via LLM for Printed Circuit Boards — Uses LLMs with constraint guidance to automatically generate PCB schematic designs. Authors, arXiv 2026. [2602.00510]
• Agentic Hardware Design Reviews — Proposes LLM-based agents to automate hardware design review processes. AllSpice Inc., arXiv 2026. [2603.15672]
• Component Centric Placement Using Deep Reinforcement Learning for PCB Design — Applies deep reinforcement learning for component placement optimization in PCB layout. Authors, arXiv 2026. [2602.23540]
• AMSnet 2.0: A Large AMS Database with AI Segmentation for Net Detection — Provides a large analog/mixed-signal database with AI-driven segmentation for net detection. Yichen Shi, Zhuofu Tao, Yuhao Gao et al., LAD 2025. [2505.09155]
• SkeySpot: Automating Service Key Detection for Digital Electrical Layout Plans in the Construction Industry — Automates service key detection in digital electrical layout plans for construction. Dhruv Dosi, Rohit Meena, Param Rajpura et al., IEEE SMC 2025. [2508.10449]
• AMSnet: A Netlist Dataset for AMS Circuits — Introduces a large-scale netlist dataset for analog and mixed-signal circuit design automation. Zhuofu Tao, Yichen Shi, Yiru Huo et al., arXiv 2024. [2405.09045]
• PCBDet — Proposes an efficient edge-deployable deep neural network for automatic PCB component detection. Brian Li, Steven Palayew, Francis Li et al., arXiv 2023. [2301.09268]
• Hand-Drawn Electrical Circuit Recognition — Recognizes hand-drawn electrical circuit diagrams using object detection and node recognition. Rachala Rohith Reddy, Mahesh Raveendranatha Panicker, arXiv 2021. [2106.11559]
• SchGen: PCB Schematic Generation with Semantic-Grounded Code Representations — Generates PCB schematics from semantic-grounded code representations using generative AI. Luo et al., arXiv 2026. [2605.30345]

Architectural Floor Plan Analysis

• PolarSym: Polar Geometry-aware Attention for CAD Floorplan Parsing — Models direction and distance in polar coordinates to improve geometrically consistent parsing of CAD floor plans. Kerui Chen, Yiqing Wang, Kangzhou Xin et al., arXiv 2026. [2608.11793]
• Raster2Seq: Polygon Sequence Generation for Floorplan Reconstruction — Generates polygon sequences from raster floor plan images for vectorized reconstruction. Authors, arXiv 2026. [2602.09016]
• A Fully Automated Hybrid Learning Scan-to-BIM Pipeline with Integrated Topology Refinement — Automates scan-to-BIM conversion using hybrid learning with topology-aware refinement. Authors, arXiv 2026. [2604.24311]
• MitUNet: Enhancing Floor Plan Recognition using a Hybrid Mix-Transformer and U-Net Architecture — Combines Mix-Transformer and U-Net for improved floor plan element recognition. Dmitriy Parashchuk, Alexey Kaspshitskiy, Yuriy Karyakin, arXiv 2025. [2512.02413]
• SAM-Guided Floorplan Reconstruction with Semantic-Geometric Fusion — Leverages Segment Anything Model with semantic-geometric fusion for floor plan reconstruction. Hanfu Ye, Hanfu Wang, Yunchi Zhang et al., arXiv 2025. [2509.15750]
• A Multi-Agent Human-AI Collaborative Pipeline to Convert Hand-Drawn Floor Plans to 3D BIM — Uses multi-agent collaboration to convert hand-drawn sketches into 3D BIM models. Authors, arXiv 2025. [2510.20838]
• Graph Similarity Learning of Floor Plans — Learns graph-based representations to measure structural similarity between floor plans. Authors, arXiv 2025. [2509.03737]
• Cloud2BIM: Open-source Automatic Pipeline for Efficient Conversion of Large-scale Point Clouds to IFC Format — Provides an open-source pipeline converting large-scale point clouds to IFC-format BIM. Authors, arXiv 2025. [2503.11498]
• WAFFLE: Multimodal Floorplan Understanding in the Wild — Proposes a multimodal benchmark for diverse real-world floor plan understanding tasks. Keren Ganon, Morris Alper, Rachel Mikulinsky et al., WACV 2025. [2412.00955]
• Offset-Guided Attention Network for Room-Level Aware Floor Plan Segmentation — Proposes an offset-guided attention mechanism for room-level segmentation of architectural floor plans. Zhangyu Wang, Ningyuan Sun, arXiv 2022. [2210.17411]
• Room Classification on Floor Plan Graphs using Graph Neural Networks — Applies graph neural networks to classify rooms from floor plan graph representations. Abhishek Paudel, Roshan Dhakal, Sakshat Bhattarai, arXiv 2021. [2108.05947]
• Deep Floor Plan Recognition Using a Multi-Task Network with Room-Boundary-Guided Attention — Introduces a multi-task network with room-boundary-guided attention for floor plan recognition. Zhiliang Zeng, Xianzhi Li, Ying Kin Yu et al., ICCV 2019. [1908.11025]

---

3D CAD Generation and Reconstruction

Methods for generating parametric 3D CAD models from various inputs including text, images, point clouds, and sketches.

Representative anchors: DeepCAD for autoregressive CAD sequence generation; SkexGen for disentangled codebook generation; recent code-generation methods for LLM-native CAD.

Autoregressive Sequence Models

• AGDC: Autoregressive Generation of Variable-Length Sequences with Joint Discrete and Continuous Spaces — Proposes autoregressive generation handling variable-length sequences in joint discrete-continuous parameter spaces. Yeonsang Shin, Insoo Kim, Bongkeun Kim et al., arXiv 2026. [2601.05680]
• Position: You Can't Manufacture a NeRF — Argues that implicit neural representations are insufficient for downstream CAD manufacturing workflows. Kimmel et al., ICML 2025.
• CAD-SIGNet: CAD Language Inference from Point Clouds using Layer-wise Sketch Instance Guided Attention — Infers CAD modeling sequences from point clouds using sketch-instance-guided attention layers. Mohammad Sadil Khan, Elona Dupont, Sk Aziz Ali et al., CVPR 2024. [2402.17678]
• FlexCAD: Unified and Versatile Controllable CAD Generation with Fine-tuned Large Language Models — Enables controllable CAD generation through fine-tuned large language models in a unified framework. Zhanwei Zhang, Shizhao Sun, Wenxiao Wang et al., ICLR 2025. [2411.05823]
• Img2CAD: Conditioned 3D CAD Model Generation from Single Image with Structured Visual Geometry — Generates 3D CAD models from single images using structured visual geometry conditioning. Authors, arXiv 2024. [2410.03417]
• Img2CAD: Reverse Engineering 3D CAD Models from Images through VLM-Assisted Conditional Factorization — Reverse engineers 3D CAD models from images via vision-language-model-assisted conditional factorization. Authors, arXiv 2024. [2408.01437]
• ContrastCAD: Contrastive Learning-based Representation Learning for Computer-Aided Design Models — Learns CAD model representations through contrastive learning for downstream design tasks. Kim et al., arXiv 2024. [2404.01645]
• Pushing Auto-regressive Models for 3D Shape Generation at Capacity and Scalability — Scales autoregressive models for 3D shape generation to higher capacity and larger datasets. Qian et al., arXiv 2024. [2402.12225]

Diffusion-Based Generation

• SketchDNN: Joint Continuous-Discrete Diffusion for CAD Sketch Generation — Proposes a joint continuous-discrete diffusion model for generating CAD sketches with mixed parameter types. Sathvik Chereddy, John Femiani, ICML 2025. [2507.11579]
• RECAD: Revisiting CAD Model Generation by Learning Raster Sketch — Generates CAD models by learning from rasterized sketch representations rather than sequential commands. Pu Li, Wenhao Zhang, Jianwei Guo et al., arXiv 2025. [2503.00928]
• Feasibility Enhancement for 3D CAD Generation — Improves physical and geometric feasibility of diffusion-generated 3D CAD models. Authors, arXiv 2025. [2505.23287]
• Physically Grounded 3D Shape Generation for Industrial Design — Generates 3D shapes grounded in physical constraints for industrial design applications. Mezghanni et al., arXiv 2025. [2512.00422]
• GenCAD: Image-Conditioned Computer-Aided Design Generation with Transformer-Based Contrastive Representation and Diffusion Priors — Generates CAD models from images using contrastive learning and diffusion priors. Md Ferdous Alam, Faez Ahmed, arXiv 2024. [2409.16294]
• Shaping Realities: Enhancing 3D Generative AI with Fabrication Constraints — Integrates fabrication constraints into 3D generative models to ensure manufacturability. Faruqi et al., arXiv 2024. [2404.10142]
• Computer-Aided Design Generation by Cascaded Discrete Diffusion Model — Generates CAD command-and-parameter sequences with a cascaded discrete diffusion model. Pan et al., arXiv 2026. [2605.05031]
• Target-Guided Bayesian Flow Networks for Quantitatively Constrained CAD Generation — Generates CAD models satisfying quantitative constraints using target-guided Bayesian flow networks. Zheng et al., arXiv 2025. [2510.25163]

LLM and VLM-Based Generation

• CIT-CAD: Constraint Intent Tree-based CAD Code Generation and Verification — Represents construction intent as explicit constraints that guide CadQuery generation, detect violations, and localize repairs. Yali Du, Hui Sun, San-Zhuo Xi et al., arXiv 2026. [2609.07434]
• VisCAD: A Foundation Model Suite with Multimodal Industrial CAD Intelligence — Maps text, renders, engineering drawings, and photographs to executable CAD programs and uses a domain harness for assembly mating and placement. Guanlin Li, Zhichao Huang, Huimu Yu et al., arXiv 2026. [2609.03811]
• ExpConCAD: Experience-Guided Text-to-CAD Generation from Shape Descriptions with Implicit Spatial Constraints — Recovers construction structure and retrieves prior design experience to complete spatial constraints omitted from text descriptions. Jingyao Liu, Jinkang Tang, Chen Huang et al., arXiv 2026. [2608.24760]
• Test-Time Scaling for CAD Generation via Verifier-Free Consensus Selection — Selects a parametric CAD program by geometric or topological agreement within a sampled candidate pool without a separate verifier. Aaron Haag, Altay Kacan, Bertram Fuchs et al., arXiv 2026. [2608.09706]
• IndustryForge-27B: A Domain-Enhanced Multimodal Foundation Model for Industrial CAD — Fine-tunes a multimodal model across CAD visual reasoning, parametric code, assemblies, and industrial software APIs. Nianchen Deng, Jiaxin Ai, Tao Hu et al., arXiv 2026. [2607.28050]
• HierCAD: Hierarchical Text-to-CAD Design via Structure Alignment and Parameter Grounding — Aligns object structures and grounds part parameters for hierarchical text-to-CAD generation. Jimin Xu, Tianbao Wang, Tao Jin et al., arXiv 2026. [2607.11339]
• Foundation Models for Automatic CAD Generation — Benchmarks foundation models and iterative critique on mechanical CAD generation tasks. J. de Curtò, Victoria Guillén, I. de Zarzà, Springer Advances in Global Applied Artificial Intelligence 2026. [2607.05573]
• Arko-T: A Foundation Model for Text-to-Structured 3D Generation — Maps text directly to executable parametric CAD programs with design-state supervision. Liang Wang, Zhaoyang Xi, Zekai Xiang et al., arXiv 2026. [2606.30429]
• Enhancing Creativity in 3D Generative Design via a TRIZ-Inspired Text-to-CAD Framework — Uses TRIZ-grounded prompting to generate editable CAD alternatives for technical contradictions. Dongeon Lee, Leekyo Jeong, Soyoung Yoo et al., arXiv 2026. [2606.21378]
• GuideCAD: A Lightweight Multimodal Framework for 3D CAD Model Generation via Prefix Embedding — Uses prefix embeddings to generate editable construction sequences from image-text inputs. Minseong Kim, Jinyeong Park, Sungho Park et al., IEEE Access 2026. [2606.07024]
• PR-CAD — Progressive refinement framework for unified controllable and faithful text-to-CAD generation using LLMs. Chen et al., ICLR 2026. [2604.19773]
• Learning Hierarchical and Geometry-Aware Graph Representations for Text-to-CAD — Learns hierarchical graph representations encoding geometry for text-driven CAD model generation. Zhang et al., ICLR 2026. [2604.10075]
• FutureCAD — Achieves high-fidelity CAD generation via LLM-driven program synthesis and text-based B-Rep primitive grounding. Jiahao Li, Qingwang Zhang, Qiuyu Chen et al., arXiv 2026. [2603.11831]
• ProCAD — Proactive agents that clarify ambiguous user intent before generating robust text-to-CAD outputs. Bo Yuan, Zelin Zhao, Petr Molodyk et al., arXiv (in review) 2026. [2602.03045]
• CAD-Tokenizer — Modality-specific tokenization enabling text-based CAD prototyping from language descriptions. Ruiyu Wang, Shizhao Sun, Weijian Ma et al., ICLR 2026. [2509.21150]
• CADSmith — Multi-agent CAD generation system with programmatic geometric validation for structural correctness. Makatura et al., arXiv 2025. [2603.26512]
• Proc3D — Procedural 3D shape generation and parametric editing driven by large language models. Li et al., arXiv 2025. [2601.12234]
• CAD-Coder — Generates CAD file code from text guidance using language models. Wang et al., arXiv 2025. [2505.08686]
• CADgpt: Harnessing Natural Language Processing for 3D Modelling to Enhance Computer-Aided Design Workflows — Leverages natural language processing to generate 3D CAD models from text instructions. Timo Kapsalis, arXiv 2024. [2401.05476]
• 3D-GPT: Procedural 3D Modeling with Large Language Models — Uses LLM agents to generate procedural 3D models via structured instructions. Sun et al., arXiv 2023. [2310.12945]
• Self-Improving CAD Generation Agents with Finite Element Analysis as Feedback — Improves learned CAD generators by using finite element analysis results as iterative feedback. Son et al., arXiv 2026. [2605.17448]
• STEP-LLM: Generating CAD STEP Models from Natural Language with Large Language Models — Generates STEP-format CAD models directly from natural language using large language models. Shi et al., arXiv 2026. [2601.12641] [Code]

Reinforcement Learning-Enhanced Generation

• RA-CAD: Learning Post-Execution Critique for State-Aware Text-to-CAD Generation — Trains an agent to turn execution feedback into outcome-aligned critiques and iterative CAD code revisions. Shuhao Yan, Changhao He, Peng Hu et al., arXiv 2026. [2608.05714]
• CME-CAD — Heterogeneous collaborative multi-expert reinforcement learning framework for CAD code generation. Zhang et al., arXiv 2025. [2512.23333]
• ReCAD — Reinforcement learning enhanced parametric CAD model generation with vision-language models. Jiahao Li, Yusheng Luo, Yunzhong Lou et al., AAAI 2026 (Oral). [2512.06328]
• CAD-RL — Multimodal chain-of-thought reinforcement learning for precise CAD code generation from intent. Ke Niu, Haiyang Yu, Zhuofan Chen et al., arXiv 2025. [2508.10118]
• Memory-Augmented Reinforcement Learning Agent for CAD Generation — Generates CAD models with a memory-augmented reinforcement learning agent for advanced manufacturing. Xiaolong et al., arXiv 2026. [2605.19748]

B-Rep and CSG Generation

• HiFi-BRep: High-Fidelity Latent Representation for Robust B-Rep Generation — Jointly predicts B-rep geometry and topology with topology-aware encoding and differentiable manifold constraints. Junhao Hou, Chenqi Luo, Pufan Wang et al., CVPR 2026. [2608.16485] [Code]
• Towards Valid B-Rep Generation: Training-Free Wireframe Anomaly Detection and Repair — Detects and repairs geometric and topological anomalies in intermediate wireframes before they produce invalid B-reps. Jingyu Wu, Youcheng Cai, Tengyu Luo et al., arXiv 2026. [2608.04955]
• TG-Diff: Coupling Discrete Topology Diffusion and Topology-conditioned Geometry Diffusions for B-Rep Generation — Couples surface-adjacency diffusion with topology-conditioned parametric surface generation. MingZe Sun, Haiyong Jiang, Bingchen Yang et al., arXiv 2026. [2607.21928]
• Autoregressive B-Rep Shape Generation with Parametric Surfaces — Generates B-reps with native surface types and continuous parameters before topology recovery. Dafei Qin, Rui Xu, Zeyu Shen et al., SIGGRAPH 2026. [2607.17093]
• HiDiGen — Hierarchical diffusion model for B-Rep generation with explicit topological constraints. Shurui Liu, Weide Chen, Ancong Wu, arXiv 2026. [2604.02847]
• BrepARG — Autoregressive B-Rep generation using a holistic token sequence representation. Jiahao Li, Yunpeng Bai, Yongkang Dai et al., CVPR 2026. [2601.16771]
• Flatten The Complex — Joint B-Rep generation via compositional k-cell particles flattening complex topology. Junran Lu, Yuanqi Li, Hengji Li et al., arXiv 2026. [2601.17733]
• Topology-First B-Rep Meshing — Prioritizes topological structure before geometry in B-Rep mesh generation. Zhou et al., arXiv 2026. [2604.02141]
• DTGBrepGen — Decouples topology and geometry into separate stages for B-Rep generation. Jing Li, Yihang Fu, Falai Chen, arXiv 2025. [2503.13110]
• AutoBrep — Autoregressive B-Rep generation unifying topology and geometry in a single model. Xiang Xu, Pradeep Kumar Jayaraman, Joseph G. Lambourne et al., SIGGRAPH Asia 2025. [2512.03018]
• HoLa — B-Rep generation using a holistic latent representation capturing full solid structure. Yilin Liu, Duoteng Xu, Xingyao Yu et al., SIGGRAPH 2025. [2504.14257]
• BrepGPT — Autoregressive B-Rep generation leveraging Voronoi half-patch decomposition. Pu Li, Wenhao Zhang, Weize Quan et al., arXiv 2025. [2511.22171]
• A Unified Differentiable Boolean Operator with Fuzzy Logic — Proposes a differentiable Boolean operator using fuzzy logic for end-to-end CSG optimization. Hsueh-Ti Derek Liu, Maneesh Agrawala, Cem Yuksel et al., SIGGRAPH 2024. [2407.10954]
• SolidGen: An Autoregressive Model for Direct B-rep Synthesis — Generates B-rep CAD solids directly via an autoregressive transformer without intermediate representations. Pradeep Kumar Jayaraman, Joseph G. Lambourne, Nishkrit Desai et al., TMLR 2023. [2203.13944]
• Implicit Conversion of Manifold B-Rep Solids by Neural Halfspace Representation — Converts manifold B-rep solids to implicit fields using learned neural halfspace representations. Hao-Xiang Guo, Yang Liu, Hao Pan et al., arXiv 2022. [2209.10191]
• CSG-Stump: A Learning Friendly CSG-Like Representation for Interpretable Shape Parsing — Introduces a flat, learning-friendly CSG representation for interpretable 3D shape parsing. Daxuan Ren, Jianmin Zheng, Jianfei Cai et al., ICCV 2021. [2108.11305]
• Learning Compact CAD Shapes with Adaptive Primitive Assembly — Learns compact shape representations by adaptively assembling geometric primitives into CAD models. Fenggen Yu, Zhiqin Chen, Manyi Li et al., arXiv 2021. [2104.05652]
• CSGNet: Neural Shape Parsers for Constructive Solid Geometry — Parses 3D shapes into CSG program sequences using a neural network. Gopal Sharma, Rishabh Goyal, Difan Liu et al., arXiv 2019. [1912.11393]

Point Cloud to CAD

• CADENA: Stepwise CAD Reverse Engineering — Reconstructs a mesh as an editable CAD program one operation at a time while comparing each intermediate geometry with the target. Soslan Kabisov, Gennadiy Savrasov, Maksim Elistratov et al., arXiv 2026. [2608.00799] [Code]
• CADReasoner — Iterative program editing approach for CAD reverse engineering from point clouds. Soslan Kabisov, Vsevolod Kirichuk, Andrey Volkov et al., CVPR 2026. [2603.29847]
• Fast Curvature Regularization of Neural SDFs for CAD Models — Accelerates curvature regularization of neural signed distance fields for CAD geometry. Kang et al., arXiv 2025. [2506.16627]
• Point2Primitive — Reconstructs CAD models from point clouds by directly predicting geometric primitives. Xinzhu Ma, Cheng Wang, Chen Tang et al., arXiv 2025. [2505.02043]
• NeurCADRecon — Reconstructs CAD surfaces via neural representation enforcing zero Gaussian curvature constraints. Kang et al., SIGGRAPH 2024. [2404.13420]
• TransCAD — Hierarchical transformer that infers CAD modeling sequences from point clouds. Elona Dupont, Kseniya Cherenkova, Dimitrios Mallis et al., arXiv 2024. [2407.12702]
• CAD-Recode — Reverse engineers parametric CAD code from point cloud inputs. Danila Rukhovich, Elona Dupont, Dimitrios Mallis et al., arXiv 2024. [2412.14042]
• PS-CAD — Leverages local geometry prompting and selection guidance for CAD reconstruction. Authors, ACM TOG 2024. [2405.15188]
• P2CADNet — End-to-end network that reconstructs featured CAD models from point clouds. Zhang et al., arXiv 2023. [2310.02638]
• Point2CAD: Reverse Engineering CAD Models from 3D Point Clouds — Reconstructs parametric CAD models from raw 3D point clouds via surface fitting and topology recovery. Yujia Liu, Anton Obukhov, Jan Dirk Wegner et al., arXiv 2023. [2312.04962]
• ComplexGen: CAD Reconstruction by B-Rep Chain Complex Generation — Generates B-Rep CAD models from point clouds by predicting vertices, edges, and faces as chain complexes. Haoxiang Guo, Shilin Liu, Hao Pan et al., SIGGRAPH 2022. [2205.14573]
• CADFit: Precise Mesh-to-CAD Program Generation with Hybrid Optimization — Recovers precise parametric CAD programs from meshes or point clouds via hybrid optimization. Nehme et al., arXiv 2026. [2605.01171] [Code]
• Extrusion Segmentation Strategy to Improve CAD Reconstruction from Point Cloud — Improves CAD reconstruction from point clouds with an extrusion-segmentation strategy. Harb et al., arXiv 2026. [2605.08971]
• Scheduling the Off-Diagonal Weingarten Loss of Neural SDFs for CAD Models — Schedules an off-diagonal Weingarten loss to improve neural SDF reconstruction of CAD models. Yin et al., arXiv 2025. [2511.03147]

Image to CAD

• RealCAD: Towards Real-World Image-to-CAD Reconstruction under Domain Shift and Parameter Bias — Reconstructs editable CAD command sequences from real photographs while correcting parameter bias, with the paired OpenRealCAD dataset. Yihe Sun, Ziyu Lu, Kaihua Tang et al., arXiv 2026. [2608.30617] [Code]
• IterCAD: Iterative Program Repair for CAD Code Generation from Orthographic Views — Generates parametric CAD code from dimensioned orthographic drawings through repeated visual comparison and program repair. Yuchuan Wu, Ke Niu, Haiyang Yu et al., ACM MM 2026. [2608.24020]
• Spline-Based Boundary Representations for Sparse View Reconstruction and Simulation Using Isogeometric Analysis — Reconstructs watertight multi-patch B-spline boundary representations from sparse images for CAD and simulation workflows. Davor Dobrota, Vsevolod Skorokhodov, Chenghao Xu et al., arXiv 2026. [2607.26234]
• Ortho2CAD: 3D CAD generation from orthographic drawings using vision language models — Converts raster orthographic drawings into editable CadQuery code using supervised fine-tuning and geometry-grounded reinforcement learning. Aditya Joglekar, Amit Regmi, Kenji Shimada et al., arXiv 2026. [2607.08891]
• SOV-CAD: Stepwise Orthographic Views Guided CAD Modeling Sequence Reconstruction — Reconstructs CAD sequences with stepwise orthographic feedback and offline reinforcement learning. Zhaopeng Feng, Chen Zhi, Xuhong Zhang et al., arXiv 2026. [2607.04119]
• GIFT: Bootstrapping Image-to-CAD Program Synthesis via Geometric Feedback — Converts kernel feedback from successful and near-miss programs into image-to-CAD training data. Giorgio Giannone, Anna Clare Doris, Amin Heyrani Nobari et al., arXiv 2026. [2603.27448]
• BrepGaussian — Reconstructs B-rep CAD models from multi-view images using Gaussian splatting representations. Wang et al., arXiv 2025. [2602.21105]
• ProcGen3D — Learns neural procedural graph representations for image-to-3D reconstruction. Gao et al., arXiv 2025. [2511.07142]
• GACO-CAD — Generates geometry-augmented and conciseness-optimized CAD models from a single image. Wang et al., arXiv 2025. [2510.17157]
• View2CAD — Reconstructs view-centric CAD models from single RGB-D scans. Noeckel et al., arXiv 2025. [2504.04000]
• CAD Sequence and Knowledge Inference — Infers CAD modeling sequences and design knowledge from product images. Authors, arXiv 2025. [2501.04928]
• DiffCAD — Performs weakly-supervised probabilistic CAD model retrieval and alignment from an RGB image. Gao et al., SIGGRAPH 2024. [2311.18610]
• Multi-View to CAD — Creates CAD models from multi-view images via geometric reconstruction. Kniaz et al., arXiv 2023. [2309.13281]
• Img2CADSeq: Image-to-CAD Generation via Sequence-Based Diffusion — Reconstructs high-quality B-rep CAD from single-view images via sequence-based diffusion. Tan et al., SIGGRAPH 2026. [2605.13293] [Code]

Sketch to CAD

• Encoded but Not Actionable: Auditing the Decode-Generate-Steer Gap in Frozen LLMs for Geometric Constraints — Uses parametric sketch constraints to separate what frozen LLMs encode from what they can generate, influence, or steer. Man Liang, Xinzhao Cheng, Faizan Wajid, arXiv 2026. [2608.17843]
• Learning Multimodal Feature-Enhanced Diffusion Models for Zero-Shot Sketch-Based 3D Shape Retrieval — Combines multimodal features with diffusion models for zero-shot 3D shape retrieval from sketches. Authors, arXiv 2026. [2604.19135]
• AutoConstrain: Aligning Constraint Generation with Design Intent in Parametric CAD — Generates geometric constraints aligned with designer intent for parametric CAD sketches. Casey et al., ICCV 2025. [2504.13178]
• Robust Self-Supervised CAD Reconstruction from Three Orthographic Views Using 3D Gaussian Splatting — Reconstructs CAD models from three orthographic views via self-supervised 3D Gaussian splatting. Zhou et al., arXiv 2025. [2503.05161]
• Efficient CAD Parametric Primitive Analysis with Progressive Hierarchical Tuning — Analyzes parametric primitives in CAD sketches using progressive hierarchical tuning for efficiency. Authors, arXiv 2025. [2503.18147]
• Sketch-Driven 3D Model Generation — Generates 3D models directly from freehand sketches as input. Authors, arXiv 2025. [2505.04185]
• DAVINCI: A Single-Stage Architecture for Constrained CAD Sketch Inference — Proposes a single-stage architecture that jointly infers primitives and constraints in CAD sketches. Mallis et al., BMVC 2024. [2410.22857]
• Parametric Primitive Analysis of CAD Sketches with Vision Transformer — Applies vision transformers to detect and parameterize geometric primitives in CAD sketches. Li et al., arXiv 2024. [2407.00410]
• PICASSO: A Feed-Forward Framework for Parametric Inference of CAD Sketches via Rendering Self-Supervision — Feed-forward network for parametric CAD sketch inference supervised through differentiable rendering. Karadeniz, Mallis, Mejri et al., WACV 2025. [2407.13394]
• 3D CAD Model Reconstruction from 2D Sketch using Visual Transformer and Rhino Grasshopper — Reconstructs 3D CAD models from 2D sketches using visual transformers integrated with Rhino Grasshopper. Authors, arXiv 2023. [2309.16850]
• 3D Modeling from Free-hand Sketches with View- and Structural-Aware Adversarial Training — Generates 3D models from freehand sketches via view-aware and structure-aware adversarial training. Authors, arXiv 2023. [2312.04435]
• Engineering Sketch Generation for Computer-Aided Design — Generates parametric engineering sketches for CAD using a generative model over constraint graphs. Willis et al., arXiv 2021. [2104.09621]
• Reconstruction of a 3D Wireframe from a Single Line Drawing via Generative Depth Estimation — Reconstructs 3D wireframes from single freehand line drawings using generative depth estimation. Cao and Lipson, arXiv 2026. [2604.13549]

CAD Editing

• TraceCAD: Trace-Guided Repair for Agentic CAD Generation — Preserves requirements, modeling steps, failures, and repair outcomes to localize and validate bounded CAD program edits. Fengxiao Fan, Jingzhe Ni, Fan Sang et al., arXiv 2026. [2608.03062]
• ArtisanCAD: An Industrial-Level CAD Agent with Expert-Grounded Knowledge Distillation — Distills expert CATIA workflows into an agent that produces editable native B-reps. Yunhan Xu, Qifeng Wu, Xunjin Li et al., arXiv 2026. [2607.05750]
• IterCAD: An Iterative Multimodal Agent for Visually-Grounded CAD Generation and Editing — Closes the loop between multimodal requests, executable CAD, visual feedback, and editing. Tao Hu, Jiaxin Ai, Licheng Wen et al., arXiv 2026. [2606.13368]
• CAD-Editor — A locate-then-infill framework with automated training data synthesis for text-based CAD editing. Li et al., arXiv 2025. [2502.03997]
• BRepLer — Language-guided editing of boundary representation CAD models via natural language instructions. Liu et al., arXiv 2025. [2508.10201]
• GenPara — Infers users' regions of interest with text-conditional shape parameters for 3D design editing. Li et al., arXiv 2025. [2503.14096]
• CADMorph — Geometry-driven parametric CAD editing via a plan-generate-verify loop. Ma et al., NeurIPS 2025. [2512.11480]
• ParSEL — Parameterized shape editing through natural language instructions. Huang et al., arXiv 2024. [2405.20319]
• Zero-shot CAD Program Re-Parameterization — Enables interactive manipulation of CAD programs without task-specific training. Cascaval et al., arXiv 2023. [2306.03217]
• Differentiable 3D CAD Programs — Proposes differentiable CAD programs enabling bidirectional editing between geometry and code. Cascaval et al., arXiv 2021. [2110.01182]

Assembly Generation

• ASSEMCAD: Production-Ready CAD Assembly Generation from Natural Language — Builds verifiable B-rep assemblies from typed parts, geometric ports, and executable mates. Yurui Dong, Shu Zou, Siqi Li et al., arXiv 2026. [2607.05123]
• Embodied CAD: Solver-Grounded LLM Agents for Parametric B-Rep Assembly Modeling — Uses typed CAD skills and exact-kernel feedback for editable B-rep assembly modeling. Fumin Liu, Haoyu Zhou, Fei Hao et al., arXiv 2026. [2606.31252]
• AADvark: Agent-Aided Design for Dynamic CAD Models — Uses LLM agents to assist designers in creating and modifying dynamic parametric CAD assemblies. Mitch Adler, Matthew Russo, Michael Cafarella, CAIS 2026. [2604.15184]
• Error Notebook-Guided, Training-Free Part Retrieval in 3D CAD Assemblies via Vision-Language Models — Leverages vision-language models with error notebooks for training-free part retrieval in CAD assemblies. Tan et al., ICLR 2026. [2509.01350]
• CADKnitter: Compositional CAD Generation from Text and Geometry Guidance — Generates compositional CAD models guided by both text descriptions and geometric constraints. Tri Le, Khang Nguyen, Baoru Huang et al., arXiv 2025. [2512.11199]
• Human-Crafted 3D Primitive Assembly Generation with Auto-Regressive Transformer — Generates human-style 3D primitive assemblies using an auto-regressive transformer architecture. Authors, arXiv 2025. [2505.04622]
• Diverse Part Synthesis for 3D Shape Creation — Synthesizes diverse interchangeable parts to enable varied 3D shape creation. Koo et al., arXiv 2024. [2401.09384]
• Category-Level Multi-Part Multi-Joint 3D Shape Assembly — Assembles 3D shapes from multiple parts with multiple joint connections at category level. Li et al., CVPR 2023. [2303.06163]
• Unsupervised 3D Shape Reconstruction by Part Retrieval and Assembly — Reconstructs 3D shapes without supervision by retrieving and assembling compatible parts. Authors, arXiv 2023. [2303.01999]
• JoinABLe: Learning Bottom-up Assembly of Parametric CAD Joints — Learns to predict joint connections for bottom-up assembly of parametric CAD models. Karl D.D. Willis, Pradeep Kumar Jayaraman, Hang Chu et al., CVPR 2022. [2111.12772]

Shape Programs and Procedural Generation

• PyTorchGeoNodes: Enabling Differentiable Shape Programs for 3D Shape Reconstruction — Integrates Blender geometry nodes into PyTorch for differentiable procedural shape fitting. Stekovic et al., CVPR 2025. [2404.10620]
• Example-driven Visual Program Learning for Generating 3D Object Arrangements — Synthesizes visual programs from examples to generate plausible 3D scene arrangements. Hu et al., 3DV 2025 (Oral). [2408.02211]
• GEMA: Generative Medial Abstractions for 3D Shape Synthesis — Generates 3D shapes via medial axis-based structural abstractions for diverse synthesis. Stekovic et al., arXiv 2024. [2402.16994]
• Discovering Abstractions for Visual Programs from Unstructured Primitives — Learns reusable program abstractions from unstructured geometric primitives for shape modeling. Bowers et al., arXiv 2023. [2305.05661]
• Learning to Infer 3D Shape Programs with Differentiable Renderer — Infers procedural shape programs from images using a differentiable rendering loss. Liu et al., arXiv 2022. [2206.12675]
• Interpretable Shape Programs — Recovers human-readable constructive shape programs that explain 3D geometry. Jones et al., arXiv 2022. [2212.11715]
• Learning to Infer and Execute 3D Shape Programs — Jointly learns to infer and execute programs that reconstruct 3D shapes from primitives. Tian et al., ICLR 2019. [1901.02875]
• Draw it like Euclid: Teaching Transformer Models to Generate CAD Profiles Using Ruler and Compass Construction Steps — Teaches transformers to generate CAD profiles via sequences of ruler-and-compass geometric constructions. Li et al., arXiv 2026. [2601.09428]

Shape Completion

• 3D Shape Completion with Latent Diffusion Models — Leverages latent diffusion models to complete partial 3D shapes from incomplete inputs. Authors, arXiv 2024. [2403.12470]
• Partial Object Completion with SDF Latent Transformers — Uses signed distance field latent transformers to complete partially observed 3D objects. Authors, arXiv 2024. [2411.05419]
• A Spatial-Aware Generative Model for 3D Shape Completion, Reconstruction, and Generation — Proposes a spatial-aware generative model unifying shape completion, reconstruction, and generation. Authors, arXiv 2024. [2403.18241]

NURBS and Surface Modeling

• Constraint-driven Optimization and Parametrization of Industrial NURBS Geometries via Neural Deformation Field — Optimizes multi-patch industrial NURBS through differentiable, constraint-aware control-point deformation. Federico Tamburlin, Giovanni Canali, Giuseppe Alessio D'Inverno et al., arXiv 2026. [2606.07198]
• Flexible Neural Surface Parameterization — Proposes a neural approach for flexible and adaptive parameterization of freeform surfaces. Authors, arXiv 2025. [2504.19210]
• Neural Parametric Surfaces for Shape Modeling — Learns neural parametric surface representations for reconstructing and modeling complex 3D shapes. Mehta et al., arXiv 2023. [2309.09911]
• NURBGen: High-Fidelity Text-to-CAD Generation through LLM-Driven NURBS Modeling — Generates editable high-fidelity CAD from text by having an LLM drive NURBS surface modeling. Usama et al., AAAI 2026. [2511.06194] [Code]

Multi-Modal CAD Generation

• MIRAGE-CAD: Construction-Mediated Multimodal Generation of Executable CAD Programs — Converts text, images, point clouds, or STEP/B-Rep geometry into construction plans and OpenCASCADE-executable Python CAD programs. Jizong Zhan, arXiv 2026. [2608.28669] [Code]
• Captioning and Generating 3D Content via Multi-modal Large Language Models — Leverages multi-modal LLMs to jointly caption and generate 3D content. Authors, arXiv 2026. [2601.21798]
• Text-Image Conditioned 3D Generation — Generates 3D assets conditioned on both text and image inputs. Authors, arXiv 2026. [2603.21295]
• Omni123: Unified Native 3D Generation and Editing within a Multimodal Framework — Unifies 3D generation and editing natively within a single multimodal framework. Authors, arXiv 2026. [2604.02289]
• Collaborative Multi-Modal Coding for High-Quality 3D Generation — Uses collaborative multi-modal coding strategies to produce high-quality 3D outputs. Authors, arXiv 2025. [2508.15228]
• Idea23D: Collaborative LMM Agents Enable 3D Model Generation from Interleaved Multimodal Inputs — Employs collaborative LMM agents to generate 3D models from interleaved multimodal inputs. Junhao Chen et al., arXiv 2024. [2404.04363]

Procedural and Constraint-Based Generation

• Procedural Material Generation with Large Vision-Language Models — Generates procedural material graphs from text or image inputs using vision-language models. Authors, arXiv 2025. [2501.18623]
• FeaGPT: An End-to-End Agentic AI for Finite Element Analysis — Proposes an agentic AI system that automates the full finite element analysis workflow. Authors, arXiv 2025. [2510.21993]
• Leveraging Automatic CAD Annotations for Supervised Learning in 3D Scene Understanding — Uses automatically generated CAD annotations to train supervised models for 3D scene understanding. Authors, arXiv 2025. [2504.13580]
• A Software Engineering-Inspired Shape Grammar for Durand's Plates — Applies software engineering principles to formalize shape grammars for architectural plate designs. Authors, arXiv 2024. [2404.14448]
• A Review on Geometric Constraint Solving — Surveys methods and algorithms for solving geometric constraint systems in CAD. Gao et al., ASME JCISE 2022. [2202.13795]

---

CAD Understanding and Retrieval

Representation learning, feature recognition, retrieval, and semantic understanding of CAD models.

Representative anchors: BRepNet and UV-Net for boundary representation learning; SketchGraphs for relational geometry modeling.

B-Rep Representation Learning

• Learn the Solid, Not the File: Canonical Inputs for Neural Networks on CAD Boundary Representations — Builds a solid-derived region graph that is invariant to B-rep repartitioning, kernel round-trips, and rigid motions. Heinrich Jiang, Hager Yasser Mohamed, Alexander Hitt et al., arXiv 2026. [2609.11573]
• Masked Topology Modeling for Self-Supervised Learning on Parametric CAD — Pretrains B-rep encoders by reconstructing masked adjacency, convexity, and curve topology. Heinrich Jiang, Jennifer Jang, arXiv 2026. [2607.20642]
• Pointer-CAD v2: Plan-Then-Construct CAD Generation with Dimension-Aware Parametric Precision — Separates planning from construction and preserves metric dimensions for precise parametric CAD generation. Dacheng Qi, Chenyu Wang, Jingwei Xu et al., arXiv 2026. [2606.29301]
• BRepMAE: Self-Supervised Masked BRep Autoencoders for Machining Feature Recognition — Pre-trains masked autoencoders on B-Rep data for self-supervised machining feature recognition. Can Yao, Kang Wu, Zuheng Zheng et al., arXiv 2026. [2602.22701]
• Boundary and Shape Representation Alignment via Self-Supervised Transformers — Aligns boundary and shape representations through self-supervised transformer learning. Authors, arXiv 2026. [2602.07429]
• MiCADangelo: Fine-Grained Reconstruction of Constrained CAD Models from 3D Scans — Reconstructs constrained CAD B-Rep models from 3D scans with fine-grained detail. Milin Kodnongbua, Benjamin Jones, Adriana Schulz et al., NeurIPS 2025. [2510.23429]

Multi-Modal CAD Representations

• BRepCLIP: Contrastive Multimodal Pretraining on BRep Primitives for CAD Understanding — Aligns native B-rep face and edge tokens with text and image embeddings. Muhammad Usama, Didier Stricker, Mohammad Sadil Khan et al., arXiv 2026. [2606.05515]
• CADCrafter — Generates parametric CAD models from unconstrained single-view images via a multi-stage pipeline. Yunlong Chen, Xiang Xu, Ganzhangqin Yuan et al., CVPR 2025. [2504.04753]
• BrepLLM — Enables large language models to natively understand boundary representation CAD geometry. Liyuan Deng, Hao Guo, Yunpeng Bai et al., arXiv 2025. [2512.16413]
• TAMM — Learns multi-modal 3D shape representations via three lightweight adapters for different modalities. Zhihao Zhang, Shengcao Cao, Yu-Xiong Wang, CVPR 2024. [2402.18490]
• ULIP-2 — Scales multimodal pre-training for 3D understanding using automatically generated language descriptions. Le Xue, Ning Yu, Shu Zhang et al., CVPR 2024. [2305.08275]
• LaGeM — A large geometry model for unified 3D representation learning and shape diffusion. Biao Zhang, Peter Wonka, ICLR 2025. [2410.01295]
• OpenShape — Scales 3D shape representation learning toward open-world understanding with multi-modal alignment. Minghua Liu, Ruoxi Shi, Kaiming Kuang et al., NeurIPS 2023. [2305.10764]
• Uni3D — Explores unified 3D representation at scale by aligning point clouds with images and text. Junsheng Zhou, Jinsheng Wang, Baorui Ma et al., ICLR 2024. [2310.06773]
• ULIP — Learns a unified representation aligning language, images, and point clouds for 3D understanding. Le Xue, Mingfei Gao, Chen Xing et al., CVPR 2023. [2212.05171]
• Point-Bind & Point-LLM — Aligns point clouds with multi-modal models for unified 3D understanding, generation, and instruction following. Ziyu Guo, Renrui Zhang, Xiangyang Zhu et al., arXiv 2023. [2309.00615]
• Self-Supervised Generative-Contrastive Learning of Multi-Modal Euclidean Input for 3D Shape Latent Representations — Combines generative and contrastive self-supervised learning on multi-modal Euclidean inputs for 3D shape embeddings. Chengzhi Wu, Julius Pfrommer, Mingyuan Zhou et al., arXiv 2023. [2301.04612]

Machining Feature Recognition

• BRepFormer: Transformer-Based B-rep Geometric Feature Recognition — Applies transformer architecture to recognize geometric features directly from B-rep CAD models. Yongkang Dai, Xiaoshui Huang, Yunpeng Bai et al., ACM ICMR 2025. [2504.07378]
• Leveraging Vision-Language Models for Manufacturing Feature Recognition in CAD Designs — Uses vision-language models to recognize manufacturing features in CAD designs without task-specific training. Lequn Chen, Muhammad Tayyab Khan, Ye Han Ng et al., arXiv 2024. [2411.02810]
• AAGNet: Automatic Machining Feature Recognition using Geometric Attributed Adjacency Graphs — Recognizes machining features automatically using graph neural networks on attributed adjacency graphs. Hongjin Wu, Ang Liu, Kai Wu, Computer-Aided Design 2024. [Paper]
• CNC-Net: Self-Supervised Learning for CNC Machining Operations — Proposes self-supervised learning to recognize CNC machining operations from 3D shapes. Mohsen Yavartanoo, Sangmin Hong, Reyhaneh Neshatavar et al., arXiv 2023. [2312.09925]
• Simplified Learning of CAD Features Leveraging a Deep Residual Autoencoder — Employs a deep residual autoencoder to simplify learning of CAD feature representations. Authors, arXiv 2022. [2202.10099]
• Geometry based Machining Feature Retrieval with Inductive Transfer Learning — Retrieves machining features from geometric data using inductive transfer learning. Sai Sree Harsha, Bharadwaj Manda, Ramanathan Muthuganapathy, arXiv 2021. [2108.11838]
• A Learning-based Approach to Feature Recognition of Engineering Shapes — Presents a learning-based method for recognizing features in engineering shape models. Authors, arXiv 2021. [2112.07962]
• FeatureNet: Machining Feature Recognition Based on 3D Convolution Neural Network — Introduces 3D CNN for machining feature recognition from volumetric CAD representations. Zhibo Zhang, Prakhar Jaiswal, Rahul Rai, CAD 2018. [Paper]
• FeatureFox: Sample-Efficient Panoptic Graph Segmentation for Machining Feature Recognition in B-Rep 3D-CAD Models — Recognizes machining features on B-rep models via sample-efficient panoptic graph segmentation. Fuchs et al., arXiv 2026. [2604.26770]

CAD Model Retrieval

• OSCAR: Open-Set CAD Retrieval from a Language Prompt and a Single Image — Retrieves CAD models from open-set databases using combined language and single-image queries. Tessa Pulli, Jean-Baptiste Weibel, Peter Hoenig et al., arXiv 2026. [2601.07333]
• CADGCL: Unsupervised Retrieval of CAD Models via Boundary Representations — Proposes unsupervised graph contrastive learning for CAD retrieval using B-Rep structures. Qin et al., The Visual Computer 2025. [Paper]
• SCA3D: Enhancing Cross-modal 3D Retrieval via 3D Shape and Caption Paired Data Augmentation — Augments paired shape-caption data to improve cross-modal 3D shape retrieval. Authors, arXiv 2025. [2502.19128]
• GC-CAD: Self-supervised Graph Neural Network for Mechanical CAD Retrieval — Introduces a self-supervised graph neural network with graph contrastive learning for CAD retrieval. Yuhan Quan, Huan Zhao, Jinfeng Yi et al., arXiv 2024. [2406.08863]
• Leveraging Cross-View Correspondence and Cross-Modal Mining for 3D Retrieval — Exploits cross-view correspondence and cross-modal mining to enhance 3D object retrieval. Authors, arXiv 2024. [2405.04103]
• FastCAD: Real-Time CAD Retrieval and Alignment from Scans and Videos — Enables real-time CAD model retrieval and alignment from RGB-D scans and video streams. Gumeli et al., arXiv 2024. [2403.15161]
• HOC-Search: Efficient CAD Model and Pose Retrieval from RGB-D Scans — Proposes hierarchical optimization for efficient joint CAD model and pose retrieval from RGB-D data. Stefan Ainetter, Sinisa Stekovic, Friedrich Fraundorfer et al., 3DV 2024. [2309.06107]
• Fine-Tuned but Zero-Shot 3D Shape Sketch View Similarity and Retrieval — Enables zero-shot sketch-based 3D shape retrieval using fine-tuned view similarity without task-specific training data. Tisse et al., arXiv 2023. [2306.08541]
• Accurate Instance-Level CAD Model Retrieval in a Large-Scale Database — Proposes instance-level CAD model retrieval with accurate alignment from large-scale shape databases. Jiaxin Wei, Haibin Huang, Chongyang Ma et al., ICCV 2023. [2207.01339]
• UVStyle-Net: Unsupervised Few-shot Learning of 3D Style Similarity Measure for B-Reps — Learns a 3D style similarity metric for B-Rep CAD models via unsupervised few-shot learning. Peter Meltzer, Hooman Shayani, Aditya Sanghi et al., ICCV 2021. [2105.02961]
• Text-to-CAD Retrieval: A Strong Baseline — Establishes a strong baseline for retrieving CAD models from natural-language queries. Pan et al., arXiv 2026. [2605.05572]

Sketch-Based Retrieval

• Multi-View Hierarchical Graph Neural Network for Sketch-Based 3D Shape Retrieval — Proposes a multi-view hierarchical graph neural network for sketch-based 3D shape retrieval. Authors, arXiv 2026. [2604.18019]
• SketchCleanNet: A Deep Learning Approach to the Enhancement and Correction of Query Sketches for a 3D CAD Model Retrieval System — Uses deep learning to clean and correct hand-drawn query sketches for improved 3D CAD retrieval. Bharadwaj Manda, Ramanathan Muthuganapathy, Computers & Graphics 2024. [2207.00732]
• CADSketchNet: An Annotated Sketch Dataset for 3D CAD Model Retrieval with Deep Neural Networks — Introduces an annotated sketch dataset and benchmarks deep neural networks for 3D CAD model retrieval. Bharadwaj Manda, Sai Sree Harsha, Subhrajit Dey et al., Computers & Graphics 2022. [2107.06212]

Shape Classification

• KDH-CAD: Knowledge-data hybrid CAD learning under data scarcity — Combines foundation-model and textbook knowledge for low-label mechanical CAD classification. Ziqin Gao, Zhijie Yang, Qiang Zou, arXiv 2026. [2606.01702]
• CSTNet: Constraint-Aware Feature Learning for Parametric Point Cloud — Learns geometric constraint features from parametric point clouds for CAD shape classification. Cheng Cheng, Changqing Zou, Ruowei Wang et al., ICCV 2025. [2411.07747]
• CAD 3D Model Classification by Graph Neural Networks: A New Approach Based on STEP Format — Classifies CAD models using graph neural networks applied directly to STEP file representations. Lorenzo Mandelli, Stefano Berretti, arXiv 2022. [2210.16815]

B-Rep Segmentation

• Geometry-Conditioned Instance Segmentation for Industrial Objects — Proposes geometry-conditioned methods for instance segmentation of industrial CAD objects. Li et al., arXiv 2026. [2602.20551]
• Repurposing 3D Generative Model for Part Segmentation — Repurposes pretrained 3D generative models to perform part segmentation tasks. Wu et al., arXiv 2026. [2603.16869]
• A2Z-10M+: Geometric Deep Learning with A-to-Z BRep Annotations for AI-Assisted CAD Modeling and Reverse Engineering — Releases more than 10 million multimodal annotations over more than one million CAD models for scan-, sketch-, text-, and B-rep-learning tasks. Pritham K. Jena, Bhavika Baburaj, Tushar Anand et al., arXiv 2026. [2603.12605]
• Joint Neural SDF Reconstruction and Semantic Segmentation for CAD Models — Jointly reconstructs signed distance fields and performs semantic segmentation on CAD models. Chen et al., arXiv 2025. [2510.03837]
• Few-shot Structure-Informed Machinery Part Segmentation with Foundation Models and Graph Neural Networks — Combines foundation models and graph neural networks for few-shot machinery part segmentation. Zhang et al., arXiv 2025. [2501.10080]
• Label-Efficient Part Segmentation — Proposes label-efficient methods to reduce annotation cost for 3D part segmentation. Liu et al., arXiv 2025. [2501.07434]
• Scan-to-BRep: BRep Boundary and Junction Detection for CAD Reverse Engineering — Detects B-Rep boundaries and junctions from 3D scans for CAD reverse engineering. Yujia Liu, Anton Obukhov, Riccardo De Lutio et al., arXiv 2024. [2409.14087]

Sequence-Based Encoding

• Unsupervised Contrastive Learning for Efficient and Robust Spectral Shape Matching — Proposes unsupervised contrastive learning to improve efficiency and robustness of spectral shape matching. Cao et al., arXiv 2026. [2603.18924]
• 3D Shape Matching: From Foundations to Open Challenges and Opportunities — Surveys 3D shape matching methods from classical foundations to modern open challenges. Eisenberger et al., arXiv 2026. [2604.01274]
• Diffusion Models for Shape Correspondence — Applies diffusion models to establish dense correspondences between 3D shapes. Attaiki et al., arXiv 2025. [2503.01845]

Assembly Understanding

• Linkify: Learning from Interface-Augmented Assembly Graphs — Learns context-aware part retrieval from corrected contact geometry in assembly graphs. Anushrut Jignasu, Daniele Grandi, arXiv 2026. [2607.01205]
• DYNAMO: Dependency-Aware Deep Learning Framework for Articulated Assembly Motion Prediction — Predicts articulated motion in mechanical assemblies using dependency-aware deep learning. Authors, arXiv 2025. [2509.12430]
• Generative 3D Part Assembly via Part-Whole-Hierarchy Message Passing — Generates 3D part assemblies using hierarchical message passing between parts and wholes. Bi'an Du, Jianxin Ma, Junyi Zhu et al., arXiv 2024. [2402.17464]
• DiffAssemble: A Unified Graph-Diffusion Model for 2D and 3D Reassembly — Unifies 2D and 3D reassembly tasks with a graph-diffusion generative model. Scarpellini et al., arXiv 2024. [2402.19302]
• What's in a Name? Evaluating Assembly-Part Semantic Knowledge in Language Models through User-Provided Names in CAD Files — Evaluates whether language models understand semantic relationships in CAD assembly-part naming. Peter Hartog, Daniele Grandi, Karl D.D. Willis et al., arXiv 2023. [2304.14275]
• Synthesizing CAD Assemblies in Fusion 360 — Proposes methods for synthesizing multi-part CAD assemblies within Fusion 360. Dominik Bauer, Nikolas Lamb, Sean Bittner, arXiv 2023. [2311.18492]
• HG-CAD: Material Prediction for Design Automation Using Graph Representation Learning — Predicts materials for CAD components using heterogeneous graph representation learning. Shijie Bian, Daniele Grandi, Pradeep Kumar Jayaraman et al., IDETC-CIE 2022 / JCISE 2024. [2209.12793]

---

Simulation and Design Optimization

AI-accelerated simulation surrogates, physics-informed methods, and topology optimization.

Representative anchors: Fourier Neural Operator for PDE surrogates; Physics-Informed Neural Networks for physics-constrained learning; TopoDiff and related models for generative topology optimization.

Neural Operators and FEA Surrogates

• NOEM: Efficient and Scalable Finite Element Method Enabled by Reusable Neural Operators — Proposes reusable neural operators to accelerate and scale finite element simulations. NOEM authors, Nature Computational Science 2026.
• A Graph Neural Network Surrogate for 3D Finite Element Modeling: Accelerated Full-Field Parameter Identification in Aluminum Alloy 6DR1 — Uses a GNN surrogate to accelerate full-field parameter identification in 3D FE models. GNN-FE authors, Modelling and Simulation in Materials Science and Engineering 2026.
• Efficient Dilated Squeeze and Excitation Neural Operator for Differential Equations — Introduces a dilated squeeze-and-excitation architecture for efficiently solving differential equations. Authors, arXiv preprint 2026. [2601.17407]
• Generative AI-Enhanced Probabilistic Multi-Fidelity Surrogate Modeling via Transfer Learning — Combines generative AI with transfer learning for probabilistic multi-fidelity surrogate models. Authors, arXiv preprint 2026. [2602.00072]
• A Unified Hierarchical Multi-Task Multi-Fidelity Framework for Data-Efficient Surrogate Modeling in Manufacturing — Presents a hierarchical multi-task multi-fidelity framework for data-efficient manufacturing surrogates. Authors, arXiv preprint 2026. [2603.09842]
• A Practical Introduction to Neural Operators in Scientific Computing — Provides a practical tutorial on neural operator methods for scientific computing applications. de Hoop et al., arXiv preprint 2025. [2503.05598]
• Equilibrium Neural Operator: A Physics-Informed Neural Operator for Multiscale Simulations — Proposes a physics-informed neural operator enforcing equilibrium constraints for multiscale simulations. EquiNO authors, arXiv preprint 2025. [2504.07976]
• A Comprehensive Evaluation of Graph Neural Networks and Physics Informed Learning for Surrogate Modelling of Finite Element Analysis — Benchmarks GNNs and physics-informed methods as surrogates for finite element analysis. Evaluation authors, arXiv preprint 2025. [2510.15750]
• An FEA Surrogate Model with Boundary Oriented Graph Embedding Approach for Rapid Design — Proposes boundary-oriented graph embeddings to accelerate FEA surrogate predictions for rapid design iteration. Xingyu Fu, Fengfeng Zhou, Dheeraj Peddireddy et al., Journal of Computational Design and Engineering 2023. [2108.13509]
• Sequential Deep Operator Networks (S-DeepONet) for Predicting Full-Field Solutions Under Time-Dependent Loads — Extends DeepONet sequentially to predict full-field structural responses under time-dependent loading conditions. Junyan He, Shashank Kushwaha, Jaewan Park et al., arXiv preprint 2023. [2306.08218]
• Reduced-order Modeling for Parameterized PDEs via Implicit Neural Representations — Uses implicit neural representations to build compact reduced-order models for parameterized PDEs. Tianshu Wen, Kookjin Lee, Youngsoo Choi, arXiv 2023. [2311.16410]
• Learning Deep Implicit Fourier Neural Operators (IFNOs) with Applications to Heterogeneous Material Modeling — Introduces implicit Fourier neural operators for learning solution maps in heterogeneous material simulations. Huaiqian You, Quinn Zhang, Colton J. Ross et al., CMAME 2022. [2203.08205]
• Toward Reusable Surrogate Models: Graph-Based Transfer Learning on Trusses — Applies graph-based transfer learning to create reusable surrogate models across different truss topologies. Whalen et al., Journal of Mechanical Design 2022. [2109.02689]
• Fourier Neural Operator for Parametric Partial Differential Equations — Learns mappings between function spaces using Fourier layers for efficient PDE solving. Li et al., ICLR 2021. [2010.08895]
• Learning nonlinear operators via DeepONet based on the universal approximation theorem of operators — Proposes DeepONet architecture for learning nonlinear operators between infinite-dimensional function spaces. Lu et al., Nature Machine Intelligence 2021. [Paper]

Computational Fluid Dynamics

• Faster by Design: Interactive Aerodynamics via Neural Surrogates Trained on Expert-Validated CFD — Enables interactive aerodynamic exploration using neural surrogates trained on expert-validated CFD simulations. Thumiger et al., arXiv 2026. [2604.18491]
• A Forward Look on AI Foundation Models in Computational Fluid Dynamics — Surveys opportunities and challenges for large-scale AI foundation models in CFD workflows. AI-CFD Foundation authors, arXiv 2025. [2511.20455]
• Accelerating Transient CFD through Machine Learning-Based Flow Initialization — Uses machine learning to generate better initial flow fields, accelerating transient CFD convergence. ML-CFD Init authors, arXiv 2025. [2503.15766]
• Physics-Constrained DeepONet for Surrogate CFD Models — Incorporates physics constraints into DeepONet architectures to build accurate surrogate CFD models. PC-DeepONet CFD authors, arXiv 2025. [2503.11196]
• Physics-Based Simulations with Masked Graph Neural Networks — Applies masked graph neural networks to improve accuracy and efficiency of physics-based simulations. Masked GNN authors, arXiv 2025. [2501.08738]
• TripOptimizer: Generative 3D Shape Optimization and Drag Prediction using Triplane VAE Networks — Proposes triplane VAE networks for joint generative 3D shape optimization and drag prediction. Thumiger et al., arXiv 2025. [2509.12224]
• Accelerating Shape Optimization by Deep Neural Networks with On-the-fly Determined Architecture — Accelerates shape optimization using DNNs whose architecture is determined dynamically during training. Kang et al., arXiv 2025. [2512.03555]
• Surrogate-Based Differentiable Pipeline for Shape Optimization — Presents a fully differentiable surrogate pipeline enabling gradient-based aerodynamic shape optimization. Muller et al., arXiv 2025. [2511.10761]
• Learning Mesh-Based Simulation with Graph Networks — Uses graph neural networks to learn physics simulations directly on mesh representations. Pfaff et al., ICLR 2021 (Outstanding Paper). [2010.03409]
• The Neural Particle Method -- An Updated Lagrangian Physics Informed Neural Network for Computational Fluid Dynamics — Combines Lagrangian particle methods with physics-informed neural networks for fluid simulation. Henning Wessels, Christian Weienfels, Peter Wriggers, arXiv preprint 2020. [2003.10208]

Physics-Informed Neural Networks

• Equation Discovery, Parametric Simulation, and Optimization Using PINN for Heat Conduction — Combines equation discovery with parametric PINN simulation for heat conduction optimization. PINN Heat Opt authors, arXiv preprint 2025. [2510.25925]
• PINNsAgent: Automated PDE Surrogation with Large Language Models — Uses LLM agents to automate physics-informed neural network construction for PDE surrogate modeling. PINNsAgent authors, ICML 2025. [2501.12053]
• Physics-Informed Diffusion Models — Integrates physical constraints into diffusion models for generating physically consistent samples. Bastek et al., ICLR 2025. [2403.14404]
• eXtended Physics-Informed Neural Network Method for Fracture Mechanics Problems — Extends PINNs with enrichment functions to model crack discontinuities in fracture mechanics. Authors, arXiv preprint 2025. [2509.13952]
• Physics-Informed Neural Operator for Learning Partial Differential Equations — Proposes a neural operator architecture incorporating physical equations to learn PDE solution mappings. Li et al., ACM/JMS 2024. [2111.03794]
• Physics-Informed Neural Networks and Extensions: A Review — Surveys PINN methodologies, training strategies, and extensions across scientific computing applications. PINN Review authors, arXiv preprint 2024. [2408.16806]
• Physics-Informed Neural Networks: From Fundamentals to Applications in Complex Systems — Reviews PINN fundamentals and their application to multiphysics and complex engineering systems. PINN Complex authors, arXiv preprint 2024. [2410.00422]
• Physics-Informed Neural Networks for Solving Thermo-Mechanics Problems of Functionally Graded Material — Applies PINNs to solve coupled thermo-mechanical problems in functionally graded materials. Thermo-PINN authors, arXiv preprint 2022. [2111.10751]
• Scientific Machine Learning Through Physics-Informed Neural Networks: Where We Are and What's Next — Comprehensive survey of PINN methods, architectures, training strategies, and open challenges. Cuomo et al., Journal of Scientific Computing 2022. [2201.05624]
• Transfer Learning Based Physics-Informed Neural Networks for Solving Inverse Problems in Engineering Structures — Applies transfer learning to PINNs for efficient inverse problem solving in structural engineering. Chen Xu, Ba Trung Cao, Yong Yuan et al., arXiv preprint 2022. [2205.07731]
• Physics-Informed Neural Networks (PINNs) for Heat Transfer Problems — Demonstrates PINNs for solving forward and inverse heat transfer problems without labeled data. Cai et al., Journal of Heat Transfer 2021. [Paper]
• Physics-informed neural networks: A deep learning framework for solving forward and inverse problems involving nonlinear partial differential equations — Introduces PINNs embedding physical laws into neural network loss functions for PDE solving. Maziar Raissi, Paris Perdikaris, George Em Karniadakis, Journal of Computational Physics 2019. [Paper]

Deep Learning for Topology Optimization

• Adversarial Agents on Topology Optimization: Understanding the Fragility and Robustness of Deep Learning-based and Physics-Based Design Models under Adversarial Perturbation — Evaluates learned topology-optimization surrogates under bounded perturbations and tests physics-in-the-loop recovery. Hoang Anh Nguyen, Yuan Hong, Hongyi Xu, arXiv 2026. [2608.22606]
• Multiscale Topology Optimization of Hyperelastic Structures Using Physics-Augmented Neural Networks — Combines physics-augmented neural networks with multiscale methods for hyperelastic topology optimization. Authors, arXiv preprint 2026. [2604.06519]
• Physics-Informed Transformer for Real-Time High-Fidelity Topology Optimization — Proposes a physics-informed transformer enabling real-time high-fidelity topology optimization predictions. Authors, arXiv preprint 2026. [2604.03522]
• Variational Quantum Latent Encoding for Topology Optimization — Leverages variational quantum circuits for latent space encoding in topology optimization. Authors, arXiv preprint 2025. [2506.17487]
• Transformer-Based Topology Optimization — Applies transformer architectures to directly predict optimized topologies from boundary conditions. Authors, arXiv preprint 2025. [2509.05800]
• A Foundation Model for Shape- and Resolution-Free Structural Topology Optimization — Introduces a foundation model that generalizes across arbitrary shapes and resolutions for topology optimization. Authors, arXiv preprint 2025. [2510.23667]
• Latent Space Diffusion for Topology Optimization — Uses diffusion models in latent space to generate diverse high-performance topology designs. Authors, arXiv preprint 2025. [2508.05624]
• Accelerating Metamaterial Topology Optimization Using Deep Super-Resolution Networks — Applies deep super-resolution networks to accelerate metamaterial topology optimization on coarse meshes. Authors, arXiv preprint 2025. [2511.04795]
• Diverse Topology Optimization Using Modulated Neural Fields (TOM) — Proposes modulated neural fields to generate diverse near-optimal topology solutions from single optimization. Authors, arXiv preprint 2025. [2502.13174]
• Accelerated Topology Optimization Design of 3D Structures Based on Deep Learning — Uses deep learning to accelerate topology optimization for 3D structural design. Xiang Cheng, Dalei Wang, Yue Pan et al., Structural and Multidisciplinary Optimization 2022. [Paper]
• TOuNN: Topology Optimization using Neural Networks — Represents topology as a continuous neural network field for gradient-based structural optimization. Chandrasekhar and Suresh, Structural and Multidisciplinary Optimization 2021. [Paper]

Generative and LLM-Driven Topology Optimization

• Multi-Material Multi-Physics Topology Optimization with Physics-Informed Gaussian Process Priors — Integrates physics-informed Gaussian process priors into multi-material, multi-physics topology optimization. Authors, arXiv preprint 2026. [2602.17783]
• Toward Large Language Model-Driven Symbolic Topology Optimisation for Rapid Structural Concept Generation in Manufacturable Design — Uses LLMs to drive symbolic topology optimization for rapid, manufacturable structural concept generation. Al Ali, Journal of Manufacturing and Materials Processing 2026. [Paper]
• Using Hand-Drawn Inputs for Diffusion-Based Topology Optimization — Leverages hand-drawn sketches as inputs to guide diffusion-based topology optimization. Authors, arXiv preprint 2026. [2603.18960]
• Guiding Topology Optimization Diffusion with Human Preferences — Incorporates human preference feedback to steer diffusion-model-based topology optimization results. Authors, arXiv preprint 2025. [2508.01589]
• Two-Stage Multiobjective Topology Optimization Method via SwinUnet with Enhanced Generalization — Proposes a two-stage SwinUnet approach for multiobjective topology optimization with improved generalization. Xiang et al., Scientific Reports 2025. [Paper]
• Diffusion Models for Topology Optimization in 3D Printing Applications — Applies diffusion models to generate topology-optimized structures tailored for 3D printing. Bekbolat et al., Journal of Applied Physics 2025. [Paper]
• Multiphysics Design Optimization via Generative Adversarial Networks — Employs GANs to accelerate multiphysics design optimization across coupled physical domains. Kazemi et al., Journal of Mechanical Design 2022. [Paper]

AI-Driven Generative Design

• A Research Prototype for Closed-Loop Generative Design of Customized Foot Orthoses via Semantic-Physics Alignment — Maps clinical intent to lattice geometry and uses a graph surrogate for rapid biomechanical feedback during orthosis design. Rui Wang, Byungwon Min, Suxing Liu, arXiv 2026. [2607.16631] [Code]
• Exploring generative design AI tools for astronomical instrumentation: a CubeSat chassis case study — Evaluates an AI- and FEA-assisted generative-design workflow against mechanical, thermal, vibration, and manufacturing constraints. Younes Chahid, Tassos Aretos, Will Cochrane et al., arXiv 2026. [2607.28217]
• Physics-in-the-Loop: A Hybrid Agentic Architecture for Validated CAD Engineering Design — Embeds engineering tools and physics checks into an iterative CAD-agent design loop. Elias Berger, Muhammad Usama, Jan Mehlstäubl et al., IJCAI-ECAI 2026 AI4Tech. [2605.19717]
• Agentic LLM Orchestration of Engineering Analysis in Product Development Design Practice — Orchestrates LLM agents to automate engineering analysis workflows in product development. Authors, arXiv preprint 2026. [2603.10249]
• SimuAgent: An LLM-Based Simulink Modeling Assistant Enhanced with Reinforcement Learning — Proposes an RL-enhanced LLM agent that assists engineers in building Simulink models. Authors, arXiv preprint 2026. [2601.05187]
• An LLM-Driven Multi-Agent Framework for Autonomous Construction of Deep Learning Surrogate Models in Subsurface Flow — Uses LLM-driven multi-agent collaboration to autonomously build surrogate models for subsurface flow. Authors, arXiv preprint 2026. [2604.11945]
• Large Language Model-Empowered Next-Generation Computer-Aided Engineering — Presents a vision for LLM-empowered CAE covering simulation, optimization, and design automation. Authors, arXiv preprint 2025. [2509.11447]
• Automating Data-Driven Modeling and Analysis for Engineering Applications using Large Language Model Agents — Automates data-driven modeling pipelines for engineering tasks via LLM agents. Authors, arXiv preprint 2025. [2510.01398]
• Large Language Models for Combinatorial Optimization of Design Structure Matrix — Applies LLMs to solve combinatorial optimization problems on design structure matrices. Authors, ICED 2025. [2506.09749]
• Bayesian and Non-Bayesian Multi-Fidelity Surrogate Models for Multi-Objective Aerodynamic Optimization Under Extreme Cost Imbalance — Compares multi-fidelity surrogate strategies for multi-objective aerodynamic shape optimization. Authors, arXiv preprint 2025. [2505.17279]
• Benchmarking Generative AI Against Bayesian Optimization for Constrained Multi-Objective Inverse Design — Benchmarks generative AI methods against Bayesian optimization for constrained multi-objective inverse design. Authors, arXiv preprint 2025. [2511.00070]
• Towards Goal, Feasibility, and Diversity-Oriented Deep Generative Models in Design — Proposes generative model frameworks that balance goal performance, feasibility, and diversity in engineering design. Lyle Regenwetter, Faez Ahmed, Journal of Mechanical Design 2022. [2206.07170]
• Evolving Through the Looking Glass: Learning Improved Search Spaces with Variational Autoencoders — Uses variational autoencoders to learn improved latent search spaces for evolutionary design optimization. Peter J. Bentley, Soo Ling Lim, Adam Gaier et al., Autodesk Research 2020. [Paper]
• COSMO-Agent: Tool-Augmented Agent for Closed-Loop Optimization, Simulation, and Modeling Orchestration — Bridges the CAD-CAE gap with a tool-augmented agent orchestrating closed-loop design, simulation, and optimization. Deng et al., arXiv 2026. [2605.20190]

---

Manufacturing-Aware Design

Design for manufacturing, additive manufacturing, assembly planning, and CAD/CAM integration.

Representative anchors: MeshCNN for mesh-based learning applicable to manufacturing; InverseCSG for reverse engineering; recent DfAM methods that integrate topology optimization with additive-manufacturing constraints.

Design for Manufacturing (DFM)

• LLM-Aided Design for Manufacturing: A Multi-Agent System for Intent-Preserving Redesign of CAD for Improved Manufacturability — Iteratively edits CadQuery programs through DFM review, compilation, and visual verification while preserving design intent. Kojo Welbeck, Xiangyu Shi, Zahra Sadeghi et al., arXiv 2026. [2609.05559]
• AIMold: An Autonomous AI-based Pipeline for Complex Mold Design — Predicts demolding directions and auxiliary components and constructs parting surfaces for complex mold assemblies. Pengyun Qiu, Shuo Wang, Zeyuan Chen et al., ECCV 2026. [2608.00800]
• Kolmogorov-Arnold Networks-Based Tolerance-Aware Manufacturability Assessment Integrating Design-for-Manufacturing Principles — Uses Kolmogorov-Arnold networks for tolerance-aware manufacturability assessment integrating DFM principles. arXiv preprint 2025. [2601.06334]
• Enhancing the Product Quality of the Injection Process Using eXplainable Artificial Intelligence — Applies explainable AI to enhance product quality in injection molding processes. arXiv preprint / Processes 2025. [2503.02338]
• Machine Learning-Based Manufacturing Cost Prediction from 2D Engineering Drawings via Geometric Features — Predicts manufacturing costs from 2D engineering drawings using geometric feature extraction and ML. arXiv preprint 2025. [2508.12440]
• Data-Driven Prediction of Casting Defects in Magnesium High-Pressure Die Casting Using Machine Learning — Predicts casting defects in magnesium high-pressure die casting using data-driven ML models. Pachandrin et al., International Journal of Metalcasting 2025. [Paper]
• An Artificial Intelligence Application for In-Process Springback Control of Sheet Metal Bending — Applies AI for real-time springback control during sheet metal bending processes. Fann et al., ASME J. Manufacturing Science and Engineering 2025. [Paper]
• DRL-Based Injection Molding Process Parameter Optimization for Adaptive and Profitable Production — Uses deep reinforcement learning to optimize injection molding parameters for adaptive production. arXiv preprint 2025. [2505.10988]
• Advancing Welding Defect Detection in Maritime Operations via Adapt-WeldNet — Proposes Adapt-WeldNet for improved welding defect detection in maritime operations. arXiv preprint 2025. [2508.00381]
• Adapting CLIP for Few-Shot Image Inspection in Manufacturing Quality Control — Adapts CLIP for few-shot visual inspection in manufacturing quality control settings. arXiv preprint 2025. [2501.12596]
• Explainable Artificial Intelligence for Manufacturing Cost Estimation and Machining Feature Visualization — Uses explainable AI to estimate manufacturing costs and visualize machining features. Soyoung Yoo, Namwoo Kang, arXiv preprint 2020. [2010.14824]
• A Machine-Learning Framework for Design for Manufacturability — Proposes a machine-learning framework to evaluate and improve design manufacturability. Aditya Balu, Sambit Ghadai, Gavin Young et al., arXiv preprint 2017. [1703.01499]
• Learning Localized Geometric Features Using 3D-CNN: An Application to Manufacturability Analysis of Drilled Holes — Applies 3D-CNNs to learn localized geometric features for manufacturability analysis of drilled holes. Aditya Balu, Sambit Ghadai, Kin Gwn Lore et al., arXiv preprint 2016. [1612.02141]

Design for Additive Manufacturing (DFAM)

• Task-Driven 3D Printability Assistance via Geometry- and Knowledge-Grounded LLM Reasoning — Grounds LLM recommendations in geometry and structured material and printer knowledge before fabrication. Zhaoda Du, Qiaojie Zheng, Xiaoli Zhang, arXiv 2026. [2608.22128]
• Towards end-to-end optimization in multimaterial 3D printing — Combines learned constitutive laws with finite-element topology and material-distribution optimization for multimaterial printing. Xue-Ling Luo, Steven Yang, Jingye Tan et al., arXiv 2026. [2607.13174] [Code]
• AgentsCAD: Automated Design for Manufacturing of FDM Parts via Multi-Agent LLM Reasoning and Geometric Feature Recognition — Detects B-rep manufacturability issues and proposes validated FDM-oriented CAD edits. Emmanuel George, Christopher Keefe, Peter Pak et al., arXiv 2026. [2607.02448]
• Discovery of Feasible 3D Printing Configurations for Metal Alloys via AI-Driven Adaptive Experimental Design — Uses AI-driven adaptive experiments to identify viable printing parameters for metal alloy additive manufacturing. Authors, arXiv preprint 2026. [2601.17587]
• Graph Neural Network-Based Topology Optimization for Self-Supporting Structures in Additive Manufacturing — Applies graph neural networks to topology optimization that ensures self-supporting structures without post-processing. Authors, arXiv preprint 2025. [2508.19169]
• Generative Artificial Intelligence in Lattice Structure Design for Additive Manufacturing: A Critical Review — Reviews generative AI methods for designing lattice structures tailored to additive manufacturing constraints. Su et al., eScience of Additive Manufacturing 2025. [Paper]
• Additive Manufacturing Processes Protocol Prediction by Artificial Intelligence using X-ray Computed Tomography Data — Predicts AM process protocols from X-ray CT scan data using artificial intelligence models. Authors, arXiv preprint 2025. [2501.14306]
• Sample-Efficient Bayesian Transfer Learning for Online Machine Parameter Optimization — Proposes Bayesian transfer learning to efficiently optimize manufacturing machine parameters with limited samples. Authors, arXiv preprint 2025. [2503.15928]
• Real-Time Decision-Making for Digital Twin in Additive Manufacturing with Model Predictive Control — Integrates model predictive control with digital twins for real-time AM process decisions. Authors, arXiv preprint 2025. [2501.07601]
• Digital Twin-Enabled Real-Time Control in Robotic Additive Manufacturing via Soft Actor-Critic RL — Applies soft actor-critic reinforcement learning for real-time control of robotic AM via digital twins. Authors, arXiv preprint 2025. [2501.18016]
• Computational, Data-Driven, and Physics-Informed ML Approaches for Microstructure Modeling in Metal AM — Surveys computational and physics-informed machine learning methods for predicting microstructure in metal AM. Authors, arXiv preprint 2025. [2505.01424]

Assembly Planning and Tolerance

• Learning-Based Strategy for Composite Robot Assembly Skill Adaptation — Learns transferable assembly skills for composite robot systems via strategy adaptation. Authors, arXiv preprint 2026. [2604.06949]
• Connector-Aware General Robotic Assembly from Instruction Manuals via Vision-Language Models — Uses vision-language models to interpret instruction manuals for connector-aware robotic assembly. Authors, arXiv preprint 2025. [2510.16344]
• AssemMate: Graph-Based LLM for Robotic Assembly Assistance — Leverages graph-based large language models to provide step-by-step robotic assembly guidance. Authors, arXiv preprint 2025. [2509.11617]
• Fabrica: Dual-Arm Assembly of General Multi-Part Objects via Integrated Planning and Learning — Integrates planning and learning for dual-arm robotic assembly of general multi-part objects. Tian et al., CoRL 2025 (Best Paper Award). [2506.05168]
• Hierarchical Hybrid Learning for Long-Horizon Contact-Rich Robotic Assembly — Proposes hierarchical hybrid learning to solve long-horizon contact-rich assembly tasks. Sun et al., CoRL 2025. [2409.16451]
• REASSEMBLE: A Multimodal Dataset for Contact-rich Robotic Assembly and Disassembly — Introduces a multimodal benchmark dataset for contact-rich robotic assembly and disassembly research. Authors, arXiv preprint 2025. [2502.05086]
• Tolerance Allocation of Complex Systems Based on Supervised Machine Learning and Adaptive Sampling — Applies supervised learning with adaptive sampling to optimize tolerance allocation in complex systems. Dantan et al., International Journal of Advanced Manufacturing Technology 2025. [Paper]
• Contact-Rich Robotic Assembly in Construction via Diffusion Policy Learning — Applies diffusion-based policy learning to contact-rich robotic assembly tasks in construction. Authors, arXiv preprint 2025. [2511.17774]
• Large-Scale Multi-Robot Assembly Planning for Autonomous Manufacturing — Proposes scalable planning methods for coordinating multiple robots in large-scale assembly tasks. Kyle Brown, Dylan M. Asmar, Mac Schwager et al., arXiv 2023. [2311.00192]
• Statistical Tolerance Allocation Design Considering Form Errors Based on Rigid Assembly Simulation and Deep Q-Network — Uses deep Q-network with rigid assembly simulation to optimize statistical tolerance allocation under form errors. Ci He, Shuyou Zhang, Lemiao Qiu et al., International Journal of Advanced Manufacturing Technology 2021. [Paper]

CAD/CAM Integration

• Design-to-Plan: A Large Language Model-Based Multi-Agent Framework for Manufacturing Process Planning from 3D CAD Models and 2D Engineering Drawings — Coordinates CAD feature recognition, drawing analysis, manufacturing knowledge, and process-planning agents in a traceable workflow. Muhammad Tayyab Khan, Lequn Chen, Wenhe Feng et al., arXiv 2026. [2608.24039]
• DeepMill: Neural Accessibility Learning for Subtractive Manufacturing — Learns tool accessibility maps via neural networks to guide subtractive milling operations. Authors, arXiv preprint 2025. [2502.06093]
• Implicit Neural Field-Based Process Planning for Multi-Axis Manufacturing — Uses implicit neural fields to automate process planning for multi-axis manufacturing. Authors, arXiv preprint 2025. [2511.17578]
• Knowledge Graph Fusion with Large Language Models for Accurate, Explainable Manufacturing Process Planning — Fuses knowledge graphs with LLMs to generate explainable manufacturing process plans. Authors, arXiv preprint 2025. [2506.13026]
• Implicit Neural Fields for Collision-Free Multi-Axis 3D Printing — Represents collision constraints as implicit neural fields for safe multi-axis 3D printing paths. Authors, arXiv preprint 2025. [2509.05345]
• A Cutting Mechanics-Based Machine Learning Modeling Method to Discover Governing Equations of Machining Dynamics — Discovers governing equations of machining dynamics using mechanics-informed machine learning. Authors, arXiv preprint 2025. [2501.14817]
• Deep Neural Implicit Representation of Accessibility for Multi-Axis Manufacturing — Encodes tool accessibility as a deep implicit representation for multi-axis machining planning. Authors, Computer-Aided Design 2024. [2409.02115]
• Automatic Feature Recognition and Dimensional Attributes Extraction From CAD Models for Hybrid Additive-Subtractive Manufacturing — Extracts manufacturing features and dimensions from CAD models for hybrid manufacturing workflows. Authors, arXiv preprint 2024. [2408.06891]
• BrepMFR: Enhancing Machining Feature Recognition in B-rep Models through Deep Learning and Domain Adaptation — Combines deep learning with domain adaptation to recognize machining features in B-rep CAD models. Zhang et al., Computer Aided Geometric Design 2024. [Paper]
• BRepGAT: Graph Neural Network to Segment Machining Feature Faces in a B-rep Model — Uses graph attention networks to segment machining feature faces directly from B-rep models. Jinwon Lee, Changmo Yeo, Sang-Uk Cheon et al., Journal of Computational Design and Engineering 2023. [Paper]
• Real-Time Tool-Path Planning Using Deep Learning for Subtractive Manufacturing — Proposes a deep learning approach for real-time tool-path planning in subtractive manufacturing. Yi-Fei Feng, Hong-Yu Ma, Li-Yong Shen et al., IEEE Transactions on Industrial Informatics 2024. [Paper]
• Large Language Models for Manufacturing — Explores applications of large language models to manufacturing processes and decision-making. Yixin Tian et al., arXiv preprint 2024. [2410.21418]
• Co-Optimization of Tool Orientations, Kinematic Redundancy, and Waypoint Timing for Robot-Assisted Manufacturing — Jointly optimizes tool orientations, redundancy resolution, and timing for robotic machining paths. Authors, arXiv preprint 2024. [2409.13448]
• Learning-Based Toolpath Planner on Diverse Graphs for 3D Printing — Learns toolpath planning strategies over graph representations for additive manufacturing. Yuming Huang et al., arXiv preprint 2024. [2408.09198]
• Reinforcement Learning-Based Cutting Parameter Dynamic Decision Method Considering Tool Wear for a Turning Machining Process — Applies reinforcement learning to dynamically adjust cutting parameters accounting for tool wear. Authors, International Journal of Precision Engineering and Manufacturing-Green Technology 2023. [Paper]
• Making Informed Decisions in Cutting Tool Maintenance in Milling — Proposes a data-driven framework for predictive cutting tool maintenance decisions in milling. Authors, arXiv preprint 2023. [2310.14629]
• Data-driven Modelling of Machine Tool Feedrate Behavior with Neural Networks — Models CNC machine tool feedrate behavior using neural networks for improved toolpath prediction. Authors, arXiv preprint 2021. [2106.09719]

---

Challenges and Future Directions

Papers analyzing open problems, technical challenges, and long-term research directions for AI in CAD.

Data and Representation Challenges

• AI+CAD Data Representation Architecture: From DeepCAD Solid Modeling to WHUCAD Industrial-Level Parametric Feature Modeling — Frames the representation gap from sketch-extrude solids to industrial parametric feature histories. Rubin Fan, Fazhi He, Yuxin Liu et al., arXiv 2026. [2606.16797]
• CADEvolve — Creates realistic CAD models through iterative program evolution strategies. Elistratov et al., arXiv 2026. [2602.16317]
• Learning From Design Procedure — Generates CAD programs by mimicking human design procedures for data augmentation. Chen et al., NeurIPS 2025 Workshop. [2603.06894]
• GenCAD-3D — Aligns multimodal latent spaces and balances synthetic datasets for CAD program generation. Yu et al., ASME J. Mechanical Design 2025. [2509.15246]
• CADmium — Fine-tunes code language models for text-driven sequential CAD design generation. Govindarajan et al., TMLR 2026. [2507.09792]

Technical Challenges

• Wrong Design Intent Is Worse Than Never Conditioning: A Derangement-Control Diagnosis of Header Conditioning in CAD Program Completion — Shows with executable assertions and a derangement control that incorrect design-intent headers can actively misdirect CAD program completion. Yang Xiao, arXiv 2026. [2607.23191] [Code]
• GeoFusion-CAD — Combines geometric state space modeling with diffusion for structure-aware parametric 3D design. Zhou et al., CVPR 2026. [2603.21978]
• ArtiCAD — Generates articulated CAD assemblies through multi-agent collaborative code generation. Shui et al., arXiv 2026. [2604.10992]

Engineering and Deployment Challenges

• neuralCAD-Edit: An Expert Benchmark for Multimodal-Instructed 3D CAD Model Editing — Introduces a benchmark for evaluating multimodal-instructed editing of 3D CAD models. Perrett et al., arXiv 2026. [2604.16170]
• Toward AI-driven Multimodal Interfaces for Industrial CAD Modeling — Explores multimodal interface design combining speech, gesture, and vision for industrial CAD workflows. Choi et al., arXiv 2025. [2503.16824]

Ecosystem Challenges

• 3DGen-Bench: Comprehensive Benchmark Suite for 3D Generative Models — Introduces a comprehensive benchmark suite for systematically evaluating 3D generative models. Zhang et al., arXiv 2025. [2503.21745]

Near-Term Directions

• BrepCoder: A Unified Multimodal Large Language Model for Multi-task B-rep Reasoning — Unifies multiple B-rep understanding and generation tasks within a single multimodal LLM. Kim et al., arXiv 2026. [2602.22284]
• cadrille: Multi-modal CAD Reconstruction with Online Reinforcement Learning — Applies online reinforcement learning to reconstruct CAD models from multi-modal inputs. Kolodiazhnyi et al., ICLR 2026 (Oral). [2505.22914]
• CAD-GPT: Synthesising CAD Construction Sequence with Spatial Reasoning-Enhanced Multimodal LLMs — Synthesizes CAD construction sequences using spatially-enhanced multimodal LLM reasoning. Wang et al., AAAI 2025. [2412.19663]
• CADDreamer: CAD Object Generation from Single-view Images — Generates editable CAD objects from a single RGB image via generative modeling. Li et al., CVPR 2025. [2502.20732]
• CAD-Assistant: Tool-Augmented VLLMs as Generic CAD Task Solvers — Augments vision-language models with CAD tools to solve diverse CAD tasks generically. Mallis et al., arXiv 2024. [2412.13810]

Mid-Term Directions

• QueryCAD: Grounded Question Answering for CAD Models — Proposes a grounded question-answering framework that reasons over 3D CAD model geometry and structure. Kienle et al., arXiv 2024. [2409.08704]

Long-Term Vision

• The Dawn of Agentic EDA: A Survey of Autonomous Digital Chip Design — Surveys autonomous agent-based approaches to electronic design automation for digital chips. Zang et al., arXiv 2025. [2512.23189]
• Intelligent CAD 2.0 — Proposes a long-term vision for next-generation intelligent computer-aided design systems. Zou et al., Visual Informatics 2024. [2410.03759]

---

Datasets and Benchmarks

Major datasets and benchmarks used across the AI for CAD research community.

Dataset Papers

• A Synthetic 3D Gear Dataset for Manufacturing Quality Inspection (MFGNet-Gear) — Releases 24,000 paired meshes and point clouds across 12 parametric gear designs and four quality classes, with a reproducible defect-generation pipeline. Ruo-Syuan Mei, Chenhui Shao, arXiv 2026. [2607.16288]
• FllumaOne: A Code-Native Multimodal CAD Dataset with Executable Programs and Kernel-Validated Feature Histories — Releases 100K executable, kernel-validated CAD programs aligned with feature histories, STEP geometry, renders, and text. Jizong Zhan, arXiv 2026. [2606.17696] [Code]
• Zero-to-CAD: Agentic Synthesis of Interpretable CAD Programs at Million-Scale Without Real Data — Synthesizes a million-scale dataset of interpretable CAD programs using agentic methods without real data. Willis et al., arXiv 2026. [2604.24479]
• STEP-Parts: Geometric Partitioning of Boundary Representations for Large-Scale CAD Processing — Releases a deterministic STEP-to-supervision toolchain and precomputed instance labels for approximately 180K DeepCAD/ABC models. Shen Fan, Mikołaj Kida, Przemyslaw Musialski, arXiv 2026. [2604.14927]
• Benchmarking Multimodal Models on Architectural and Engineering Drawings Understanding — Benchmarks multimodal models on their ability to understand architectural and engineering drawings. Zhang et al., arXiv 2026. [2601.04819]
• Geometrically Constrained Parametric History-based CAD Dataset — Introduces a CAD dataset with geometric constraints and parametric modeling history. Authors et al., arXiv 2025. [2602.19171]
• Objaverse++: Curated 3D Object Dataset with Quality Annotations — Provides a curated large-scale 3D object dataset enhanced with quality annotations. Authors et al., arXiv 2025. [2504.07334]
• An Open Large-Scale Architectural CAD Dataset and New Baseline for Panoptic Symbol Spotting — Releases a large-scale architectural CAD dataset with baselines for panoptic symbol spotting. Luo et al., NeurIPS 2025. [2503.22346]
• Enginuity: Building an Open Multi-Domain Dataset of Complex Engineering Diagrams — Builds an open multi-domain dataset of complex engineering diagrams for diagram understanding. Authors et al., NeurIPS 2025. [2601.13299]
• VideoCAD: A Large-Scale Video Dataset for Learning UI Interactions and 3D Reasoning from CAD Software — Presents a large-scale video dataset capturing UI interactions and 3D reasoning in CAD software. Lambourne et al., NeurIPS 2025. [2505.24838]
• SldprtNet: A Large-Scale Multimodal Dataset for CAD Generation in Language-Driven 3D Design — Introduces a large-scale multimodal dataset linking natural language to CAD model generation. Guo et al., arXiv 2025. [2603.13098]
• A Cascade MAR with Topology Predictor for Multimodal Conditional CAD Generation — Introduces a cascade masked autoregressive model with topology prediction for multimodal CAD generation. Wang et al., arXiv 2025. [2504.20830]
• mrCAD: Multimodal Refinement of Computer-aided Designs — Presents a dataset and benchmark for iterative multimodal refinement of CAD designs. McCarthy et al., EMNLP 2025 Findings. [2504.20294]
• Reinforcement Learning Training Gym for Revolution Involved CAD Command Sequence Generation — Provides an RL training environment for generating CAD command sequences involving revolution operations. Yin et al., arXiv 2025. [2503.18549]
• Synthetic Generation Tool of Digital Measurement Device CAD Model Datasets for Fine-tuning Large Vision-Language Models — Proposes a synthetic data generation tool for measurement device CAD models to fine-tune VLMs. Authors et al., arXiv 2025. [2508.21732]
• IEC3D-AD: A 3D Dataset of Industrial Equipment Components for Unsupervised Point Cloud Anomaly Detection — Introduces a 3D industrial equipment component dataset for unsupervised point cloud anomaly detection. Wang et al., arXiv 2025. [2511.03267]
• CAD-MLLM: Unifying Multimodality-Conditioned CAD Generation With MLLM — Unifies multimodal-conditioned CAD generation within a single multimodal large language model framework. Xu et al., arXiv 2024. [2411.04954]
• Text2CAD: Generating Sequential CAD Models from Beginner-to-Expert Level Text Prompts — Presents a dataset and method for generating sequential CAD models from varied-difficulty text prompts. Khan et al., NeurIPS 2024 Spotlight. [2409.17106]
• Slice-100K: A Multimodal Dataset for Extrusion-based 3D Printing — Provides 100K sliced 3D printing files with multimodal annotations for manufacturing research. Authors et al., NeurIPS 2024 D&B Track. [2407.04180]
• From Engineering Diagrams to Graphs — Converts engineering diagrams into structured graph representations for automated understanding. Chen et al., arXiv 2024. [2411.13929]
• Objaverse: A Universe of Annotated 3D Objects — Presents a large-scale dataset of 800K+ annotated 3D objects for vision and language tasks. Deitke et al., CVPR 2023. [2212.08051]
• Objaverse-XL: A Universe of 10M+ 3D Objects — Scales 3D object datasets to over 10 million objects sourced from diverse repositories. Deitke et al., NeurIPS 2023. [2307.05663]
• DeepPatent2: A Large-Scale Benchmarking Corpus for Technical Drawing Understanding — Offers a large-scale patent drawing benchmark for technical illustration recognition and retrieval. Kucer et al., Scientific Data 2023. [2311.04098]
• FloorPlanCAD: A Large-Scale CAD Drawing Dataset for Panoptic Symbol Spotting — Introduces a large-scale floorplan CAD dataset with panoptic-level symbol annotations. Fan et al., ICCV 2021 / TPAMI 2022. [2105.07147]
• DeepCAD: A Deep Generative Network for Computer-Aided Design Models — Introduces a dataset and generative model for CAD command sequence generation. Wu et al., ICCV 2021. [2105.09492]
• Fusion 360 Gallery: A Dataset and Environment for Programmatic CAD Construction from Human Design Sequences — Provides real human CAD design sequences from Autodesk Fusion 360 for reconstruction research. Willis et al., SIGGRAPH 2021. [2010.02392]
• AutoMate: A Dataset and Learning Approach for Automatic Mating of CAD Assemblies — Provides a dataset and methods for predicting mate constraints in CAD assemblies. Jones et al., SIGGRAPH 2021. [2105.12238]
• Synthetic 3D Data Generation Pipeline for Geometric Deep Learning in Architecture — Presents a pipeline for generating synthetic 3D architectural data for deep learning. Stojanovic et al., arXiv 2021. [2104.12564]
• SketchGraphs: A Large-Scale Dataset for Modeling Relational Geometry in Computer-Aided Design — Offers a large-scale dataset of parametric CAD sketches with geometric and constraint graphs. Seff et al., ICML 2020 Workshop. [2007.08506]
• A Large-Scale Annotated Mechanical Components Benchmark for Classification and Retrieval Tasks with Deep Neural Networks — Introduces an annotated benchmark of mechanical components for 3D classification and retrieval. Kim et al., ECCV 2020. [Paper]
• MFCAD: A Dataset of 3D CAD Models with Machining Feature Labels — Provides labeled CAD models annotated with machining feature types for recognition tasks. Cao et al., CAD Journal 2020. [Paper]
• ABC: A Big CAD Model Dataset For Geometric Deep Learning — Offers one million CAD models with ground-truth geometry for geometric deep learning. Koch et al., CVPR 2019. [1812.06216]
• ShapeNet: An Information-Rich 3D Model Repository — Provides a richly annotated large-scale repository of 3D shapes organized by WordNet taxonomy. Chang et al., arXiv 2015. [1512.03012]
• 3D ShapeNets: A Deep Representation for Volumetric Shapes — Introduces a large-scale 3D shape dataset and a deep volumetric representation for recognition. Wu et al., CVPR 2015. [1406.5670]
• SESYD: A Synthetic Document Database for Performance Evaluation — Provides synthetic engineering drawing symbols and diagrams for document recognition benchmarking. Delalandre et al., DAS 2010. [Paper]
• CADFS: A Big CAD Program Dataset and Framework for Computer-Aided Design with Large Language Models — Provides a large CAD-program dataset and framework enabling vision-language models to generate complex design histories. Pyatov et al., arXiv 2026. [2605.01925]

Evaluation Metrics

• Quantitative Evaluation of 3D Models Generation by Large Language Models — Proposes quantitative metrics to assess quality of 3D models generated by LLMs. Authors, arXiv 2025. [2509.07010]
• CAD-Judge: Toward Efficient Morphological Grading and Verification for Text-to-CAD Generation — Introduces an automated judge for morphological grading and verification of text-to-CAD outputs. Authors, arXiv 2025. [2508.04002]
• Generating CAD Code with Vision-Language Models for 3D Designs — Leverages vision-language models to generate CAD code and evaluates output fidelity. Authors, ICLR 2025. [2410.05340]
• Advancing 3D Generation Evaluation with Hierarchical Validity — Proposes a hierarchical validity framework for more fine-grained 3D generation evaluation. Authors, arXiv 2025. [2508.05609]
• CadVLM: Bridging Language and Vision in the Generation of Parametric CAD Sketches — Bridges language and vision modalities to generate and evaluate parametric CAD sketches. Authors, arXiv 2024. [2409.17457]

Benchmark Challenges

• ParamCAD-AgentBench — Releases an executable long-horizon benchmark with 2,409 kernel-validated parametric CAD models and 4,818 paired language-agent tasks across core and challenge splits. ParamCAD-AgentBench contributors, GitHub 2026. [Benchmark]
• ParaEval — Evaluates executed Rhino/Grasshopper parametric designs across runtime validity, measured geometry, visual intent, and headless structural solvability with explicit abstention for unmeasurable layers. Magnus Huber / Technical University of Munich, GitHub 2026. [Benchmark]
• PhysicsBench: A Unified Leaderboard for Generative and Predictive Models in Engineering Design and Simulation — Standardizes evaluation of geometry generation and physical prediction across CAD, CFD, and FEA tasks and data scales. Sang Won Lee, Hyogu Jeong, Namwoo Kang, arXiv 2026. [2608.24056] [Code]
• CADEngBench: It Looks Like CAD, but Does It Work? Evaluating Parametric Design, Assembly Reasoning, and Physics Simulation — Tests CAD systems through parametric perturbations, functional edits, DFM checks, matched FEA, and joint grounding. Harmanjot Singh, Abhra Dubey, Jorge Alejandro Amador Herrera, arXiv 2026. [2608.09296]
• OmniMech: All-in-one Multimodal Mechanical Benchmark for 3D Reconstruction — Pairs dimensioned engineering drawings with native CAD, STEP, B-rep, renderings, and annotations for executable reconstruction and agentic reasoning. Taiting Lu, Runze Liu, Ziwei Dong et al., arXiv 2026. [2608.05539]
• BIM-Edit: Benchmarking Large Language Models for IFC-Based Building Information Modeling — Provides 324 natural-language editing tasks over IFC building models with geometric, semantic, and topological evaluation. Bharathi Kannan Nithyanantham, Clemens Kujat, Tobias Sesterhenn et al., arXiv 2026. [2606.20146]
• UniCAD: A Unified Benchmark and Universal Model for Multi-Modal Multi-Task CAD — Unifies point, text, image, and sketch CAD reconstruction, generation, and question answering in one benchmark and model. Jingyuan Chen, Sheng Jin, Haopeng Sun et al., arXiv 2026. [2606.05058]
• CADBench: A Multimodal Benchmark for AI-Assisted CAD Program Generation — Provides 18K multimodal CAD-program tasks with geometry, execution, and program-quality metrics. Anna C. Doris, Jacob Thomas Sony, Ghadi Nehme et al., arXiv 2026. [2605.10873]
• CAD Arena: Open Benchmark for AI-Generated Parametric CAD — Provides an open platform for evaluating and comparing AI-generated parametric CAD models. CAD Arena Team, Online Platform 2025.
• State Space Model For 3D Computer-Aided Design Generative Modeling — Applies state space models to generative modeling of 3D CAD sequences. Authors, arXiv 2025. [2603.00439]
• BlenderLLM: Training Large Language Models for Computer-Aided Design with Self-improvement — Trains LLMs for CAD generation using a self-improvement learning framework. Authors, arXiv 2024. [2412.14203]
• Geometric Deep Learning for Computer-Aided Design: A Survey — Surveys geometric deep learning techniques applied to CAD representation and generation. Lambourne et al., arXiv 2024. [2402.17695]
• MUSE: Benchmarking Manufacturable, Functional, and Assemblable Text-to-CAD Generation — Benchmarks text-to-CAD on manufacturability, functionality, and assemblability for industrial product design. Dong et al., arXiv 2026. [2605.28579]
• Text2CAD-Bench: A Benchmark for LLM-based Text-to-Parametric CAD Generation — Benchmarks LLM-based generation of parametric CAD models from natural language. Wang et al., arXiv 2026. [2605.18430]
• Text-to-CAD Evaluation with CADTests — Introduces CADTests, a functional test-based evaluation protocol for text-to-CAD generation. Mallis et al., arXiv 2026. [2605.07807] [Code]
• BenchCAD: A Comprehensive, Industry-Standard Benchmark for Programmatic CAD — Benchmarks programmatic CAD code generation from visual or textual inputs against industry standards. Zhang et al., arXiv 2026. [2605.10865]

---

Tools and Platforms

Commercial CAD with AI Features

• SOLIDWORKS Design AI Virtual Companions — Embeds AURA and LEO in SOLIDWORKS for parametric CAD generation, assembly planning, legacy B-rep conversion, drawing generation, and model-error diagnosis. Dassault Systèmes, Technical Platform 2026. [Paper]
• Onshape Labs FeatureScript MCP Server — Connects AI clients to Onshape's native FeatureScript workflow to generate, execute, test, and refine reusable parametric CAD features from natural language. Onshape / PTC, Technical Platform 2026. [Paper]
• PTC Creo 13 AI Assistant — Adds an embedded conversational assistant for contextual CAD guidance and workflow support inside Creo 13. PTC, Technical Platform 2026. [Paper]
• Ansys GeomAI — Learns from reference geometries to generate new engineering concepts grounded in geometric and design constraints. Ansys / Synopsys, Technical Platform 2026. [Paper]
• Autodesk Fusion AI — Integrates natural-language assistance, editable geometry generation, automated drawings, sketch constraints, generative design, and CAM automation into Fusion. Autodesk, Technical Platform 2026. [Paper]
• CloudNC CAM Assist — Generates machining strategies, toolpaths, cutting parameters, machinability feedback, cycle-time estimates, and fixture geometry inside major CAM systems. CloudNC, Technical Platform 2026. [Paper]
• AI-Assisted Analysis and Synthesis of Engineering Systems from Multimodal Engineering Data — Proposes AI methods to analyze and synthesize engineering systems from multimodal data sources. H. Sinan Bank, Daniel R. Herber, IISE 2026. [2603.00251]
• Large Language Models for Computer-Aided Design: A Survey — Surveys applications of large language models across CAD tasks and workflows. Zhang et al., arXiv preprint 2025. [2505.08137]
• A Multidisciplinary Design and Optimization (MDO) Agent Driven by Large Language Models — Proposes an LLM-driven agent for automated multidisciplinary design optimization. Guo et al., arXiv preprint 2025. [2511.17511]
• Automated CAD Modeling Sequence Generation from Text Descriptions via Transformer-Based Large Language Models — Generates CAD modeling operation sequences from natural language descriptions using transformers. Liao et al., arXiv preprint 2025. [2505.19490]
• AI-Driven Digital Twins for Manufacturing — Reviews AI-driven digital twin approaches across hierarchical manufacturing system levels. Nguyen et al., Sensors 2025. [Paper]
• Autodesk Neural CAD — Presents 3D generative AI foundation models for design-to-make workflows. Autodesk, Technical Report 2025. [Paper]
• Siemens Design Copilot NX — Provides an AI copilot for product engineering within the NX CAD environment. Siemens Digital Industries Software, Technical Platform 2025. [Paper]
• Siemens Digital Twin Composer for Industrial Metaverse Digital Twins — Platform enabling creation of interactive digital twins for industrial metaverse applications. Siemens Digital Industries Software, Technical Platform 2026. [Paper]
• A Survey of AI Methods for Geometry Preparation and Mesh Generation in Engineering Simulation — Surveys AI techniques for automating geometry cleanup and mesh generation in CAE workflows. Steven Owen, Nathan Brown, Nikos Chrisochoides et al., arXiv 2025. [2512.23719]
• MotionAnymesh: Physics-Grounded Articulation for Simulation-Ready Digital Twins — Generates physically plausible articulated motion for mesh models to create simulation-ready digital twins. WenBo Xu, Liu Liu, Li Zhang et al., arXiv 2026. [2603.12936]
• NVIDIA Omniverse Blueprint for Real-Time Computer-Aided Engineering Digital Twins — Blueprint for building real-time physics-based digital twins integrated with CAE software. NVIDIA, Technical Platform 2024. [Paper]
• A Survey on AI-Driven Digital Twins in Industry 4.0: Smart Manufacturing and Advanced Robotics — Surveys AI-driven digital twin approaches for smart manufacturing and robotics applications. Huang et al., Sensors 2021. [Paper]

AI-Native CAD Platforms

• Backflip AI — Converts 3D scans, STL files, and meshes into editable parametric CAD models with feature trees, including an available Autodesk Fusion add-in. Backflip, Technical Platform 2026. [Paper]
• DraftAid — Automates production-ready 2D fabrication drawings from 3D CAD models while applying company templates, dimensioning rules, and drafting standards. DraftAid, Technical Platform 2026. [Paper]
• Neural Concept: Physics- and Geometry-Aware AI Design Copilot for Engineering — Accelerates engineering design with AI that understands physical constraints and 3D geometry. Neural Concept, Technical Platform 2025.
• Adam: AI-Native CAD Platform for Text-to-Parametric Design — Generates editable parametric CAD models from natural language descriptions. Adam (YC W25), Technical Platform 2025.
• Leo AI: Large Mechanical Model for CAD-Aware Engineering Assistance — Applies a domain-specific large model to assist mechanical engineers within CAD environments. Leo AI, Technical Platform 2025.
• Zoo.dev Text-to-CAD: Parametric CAD Generation via KCL Programming Language — Converts text prompts into parametric CAD geometry using a code-first modeling language. Zoo (formerly KittyCAD), Technical Platform 2024.
• MecAgent: AI Copilot for Mechanical CAD Software Automation — Automates repetitive mechanical CAD tasks through an AI-driven copilot interface. MecAgent, Technical Platform 2024.
• Luminary Cloud: Physics AI Factory for Real-Time Engineering Simulation — Delivers GPU-accelerated physics simulation for real-time engineering design exploration. Luminary Cloud, Technical Platform 2024.

Open-Source Tools and Frameworks

• build123d-mcp — Exposes iterative build123d/OpenCascade modeling, rendering, inspection, repair, measurement, and STEP/STL/SVG/DXF export to AI clients through MCP. Paul Fremantle, GitHub 2026. [Project]
• MAC (Multi-Agent CAD) — Implements a four-agent build123d pipeline with structured state transfer, executable geometry checks, repair loops, and a reproducible feature benchmark. Tsinghua University IEI Lab, GitHub 2026. [Paper]
• Text23D Mechanical — Provides a local conversational CAD workspace with CadQuery and FreeCAD backends, editable artifacts, provider adapters, and streamed agent execution. Text23D, GitHub 2026. [Paper]
• Sphaire — Runs AI-assisted parametric CAD in the browser using OpenCascade/Replicad, inspectable construction code, geometry validation, DFM checks, and local or hosted model providers. Sphaire contributors, GitHub 2026. [Paper]
• Chamfer — Provides a kernel-verified text/image-to-parametric-CAD agent harness with reproducible geometry-oracle benchmarks. SmartAI, GitHub 2026. [Paper]
• TOOLCAD — Leverages tool-using LLMs with reinforcement learning for text-to-CAD generation. Gong et al., arXiv 2026. [2604.07960]
• PLLM — Proposes pseudo-labeling large language models for CAD program synthesis. Li et al., arXiv 2026. [2602.12561]
• CADDesigner — General-purpose agent for conceptual CAD model generation from high-level design intent. Fan, Ni, Yin et al., arXiv preprint 2025. [2508.01031]
• Generative AI for CAD Automation — Leverages large language models to automate 3D modelling workflows in CAD. Kumar et al., arXiv preprint 2025. [2508.00843]
• CAD-Llama — Leverages large language models for parametric 3D CAD model generation from text. Li et al., arXiv preprint 2025. [2505.04481]
• Text-to-CAD Generation — Infuses visual feedback into large language models for text-to-CAD generation. Wang et al., arXiv preprint 2025. [2501.19054]
• Autodesk and Model Context Protocol — Makes MCP enterprise-ready for connecting AI agents to CAD design data. Autodesk, Technical Blog 2025. [Paper]
• FreecadMCP — Model Context Protocol server enabling AI-driven parametric design in FreeCAD. bonninr (open-source), GitHub 2025. [Paper]
• CAD Skills — Agent skills framework for parametric CAD generation via the build123d library. earthtojake (open-source), GitHub 2025. [Paper]
• CADAM — Open-source text-to-CAD web application powered by OpenSCAD-WASM. Adam-CAD (open-source), GitHub 2026. [Paper]
• HiCAD — Parametric 3D CAD modeling platform integrating JSCAD with multi-LLM support for AI-driven design. MrXujiang, GitHub 2025. [Paper]
• CQAsk — LLM-powered CAD generation tool that produces 3D models using CadQuery from natural language. OpenOrion, GitHub 2024. [Paper]
• multiCAD-mcp — MCP server enabling AI assistants to control AutoCAD, ZWCAD, and BricsCAD simultaneously. AnCode666, GitHub 2025. [Paper]
• CAD Agent — AI-driven CAD modeling agent with visual feedback loop using build123d and MCP. Svetlana-DAO-LLC, GitHub 2025. [Paper]
• gNucleus Text-to-CAD MCP — MCP server for generating CAD parts and assemblies from text descriptions. gNucleus, GitHub 2025. [Paper]
• cadquery-mcp-server — MCP server for generating and verifying CAD models through CadQuery with automated validation. Rishi Gundakaram, GitHub 2025. [Paper]
• FreeCAD MCP — Model Context Protocol server for FreeCAD with integrated finite element analysis support. neka-nat, GitHub 2025. [Paper]
• Dingcad — Live-reload CAD scripting environment combining ManifoldCAD geometry kernel with QuickJS runtime. yacineMTB, GitHub 2025. [Paper]
• OpenSCAD Agent — Claude Code-powered agent that generates 3D-printable designs through natural language to OpenSCAD code. iancanderson, GitHub 2025. [Paper]
• CodeToCAD — Vendor-agnostic framework enabling code-based CAD automation across multiple modeling backends. CodeToCAD, GitHub 2024. [Paper]
• Curated Code CAD — Comprehensive curated list of code-based CAD tools, languages, and frameworks. Irev-Dev, GitHub 2024. [Paper]
• The Power of Small LLMs in Geometry Generation for Physical Simulations — Demonstrates small language models can effectively generate geometry for physical simulation workflows. Ossama Shafiq, Bahman Ghiassi, Alessio Alexiadis, arXiv 2025. [2503.18178]
• DreamCAD — Scales multi-modal CAD generation using differentiable parametric surface representations. Mohammad Sadil Khan, Muhammad Usama, Rolandos Alexandros Potamias et al., arXiv 2026. [2603.05607]
• Generating Human-AI Collaborative Design Sequence for 3D Assets — Proposes differentiable operation graphs to model human-AI collaborative 3D design sequences. Xiaoyang Huang, Bingbing Ni, Wenjun Zhang, arXiv 2025. [2508.17645]
• A Solver-Aided Hierarchical Language for LLM-Driven CAD Design — Introduces a solver-aided hierarchical language enabling LLMs to produce constraint-satisfying CAD designs. Benjamin T. Jones, Felix Hahnlein, Zihan Zhang et al., arXiv 2025. [2502.09819]
• Generative Parametric Design: A Framework for Real-time Geometry Generation and On-the-fly Multiparametric Approximation — Proposes a framework for real-time parametric geometry generation with on-the-fly multiparametric approximation. Mohammed El Fallaki Idrissi, Jad Mounayer, Sebastian Rodriguez et al., arXiv 2025. [2512.11748]
• Designing a Human-AI Collaborative Ideation System for Concept Designers — Presents a human-AI collaborative system supporting concept designers during early-stage ideation. Wen-Fan Wang, Chien-Ting Lu, Nil Ponsa Campanya et al., arXiv 2025. [2502.14747]
• Query2CAD: Generating CAD models using natural language queries — Generates CAD models directly from natural language queries via a language-driven pipeline. Badagabettu et al., arXiv 2024. [2406.00144]
• Experiments on Generative AI-Powered Parametric Modeling and BIM for Architectural Design — Explores generative AI for parametric modeling and BIM in architectural design workflows. Jaechang Ko, John Ajibefun, Wei Yan, arXiv 2023. [2308.00227]

Research on AI Tools and Deployment

• LLM-based Visual Code Completion for Aerospace Geometric Design — Adds a ReAct-style Grasshopper copilot, Wingbuilder geometry library, and 18-task aerospace benchmark. Hau Kit Yong, Robert Marsh, Edmar A. Silva et al., arXiv 2026. [2606.16806]
• Supervising Ralph Wiggum: Exploring a Metacognitive Co-Regulation Agentic AI Loop for Engineering Design — Explores a metacognitive co-regulation loop for supervising agentic AI in engineering design tasks. Xu et al., arXiv 2026. [2603.24768]
• Is Academia Catching Up with Industry Demands? AI for CAE User Experience -- A Multivocal Literature Review — Multivocal literature review comparing academic and industry perspectives on AI for CAE user experience. Authors et al., arXiv 2025. [2507.16586]
• Beyond Development: Challenges in Deploying Machine Learning Models for Structural Engineering Applications — Identifies challenges in deploying ML models for real-world structural engineering applications. Zaker Esteghamati et al., arXiv 2024. [2404.12544]
• Naming the Pain in Machine Learning-Enabled Systems Engineering — Categorizes pain points encountered when integrating machine learning into systems engineering workflows. Kalinowski et al., arXiv 2024. [2406.04359]
• Towards a Framework for Deep Learning Certification in Safety-Critical Applications Using Inherently Safe Design and Run-Time Error Detection — Proposes a certification framework combining inherently safe design with runtime error detection for deep learning. Valentin, arXiv 2024. [2403.14678]
• Artificial Intelligence for Safety-Critical Systems in Industrial and Transportation Domains: A Survey — Surveys AI methods and challenges for safety-critical systems in industrial and transportation domains. Nascimento et al., ACM Computing Surveys 2023. [Paper]
• Challenges in Deploying Machine Learning: A Survey of Case Studies — Surveys real-world case studies to identify recurring challenges in deploying machine learning systems. Paleyes et al., ACM Computing Surveys 2022. [Paper]

---

Contributing

Contributions are welcome. To add a paper or resource:

1. Fork this repository
2. Add the entry in the appropriate section, following the existing format
3. Ensure the paper link is valid (use arXiv links when available)
4. Run python3 scripts/validate_catalog.py
5. Submit a pull request with a brief description

Please ensure entries are:

• Placed in the correct thematic section
• Sorted by year (newest first) within each subsection
• Formatted consistently with existing entries

For questions or suggestions, please open an issue.

Maintenance

Run the catalog validator before submitting maintenance changes:

python3 scripts/validate_catalog.py

Catalog Metadata

This repository intentionally separates three denominators:

Use these terms explicitly when citing counts. Do not collapse them into an undefined "papers" total.

For the latest full catalog-entry confidence review, see catalog_entry_audit_summary_2026-05-30.md. The subsequent incremental review is documented in catalog_increment_review_2026-09-14.md.

---

License

This project is licensed under the MIT License. See LICENSE for details.