Awesome Post-Training in Autonomous Driving Papers

A curated list of papers on post-training for end-to-end autonomous driving, the stage that refines a driving policy after imitation learning, using supervision beyond offline expert demonstrations. This includes distillation, preference-based alignment, reinforcement learning (RL), and test-time refinement.

This repository accompanies our survey:

Post-Training in End-to-End Autonomous Driving: A Unified View
Ruining Yang*, Muxing Wang*, Yixiao Chen, Tongfei Guo, Yi Xu, Can Cui, Zichong Yang, Yitian Zhang, Ziran Wang, Yun Fu, Lili Su.
(* Equal contribution. Northeastern University & Purdue University.)

---

🧭 Taxonomy

We organize post-training methods into four families by the form of supervision they use:

Family tags used below: [Distill] · [Preference] · [RL] · [Test-time] · [SFT] (continued / targeted supervised fine-tuning) · [Planner-RL] (RL post-training on non-foundation-model planners).

Each entry also carries a venue tag in front (e.g. [CVPR], [ICLR], [NeurIPS], [arXiv]) so the publication venue is visible at a glance.

---

📌 Contents

• Papers by Year
  • 2026
  • 2025
  • 2024
• Benchmarks & Simulators
• Related Surveys
• Citation
• Contributing

---

📄 Papers by Year

Papers are grouped by the year of first public (arXiv) release, and clustered by family within each year. Each entry is prefixed with a venue tag and its family tag; the published venue is also noted in italics. A paper is listed under its primary family; some methods span multiple families.

2026

• [Comm. Transp. Res.] [Distill] Found-RL: Foundation Model-Enhanced Reinforcement Learning for Autonomous Driving, Communications in Transportation Research 2026. arXiv Code

• [CVPR] [Distill] PaIR-Drive: Fine-tuning is not Enough — A Parallel Framework for Collaborative Imitation and Reinforcement Learning in End-to-End Autonomous Driving, CVPR 2026. arXiv Code

• [arXiv] [Distill] CRAFT: Counterfactual-to-Interactive Reinforcement Fine-Tuning for Driving Policies, arXiv 2026. arXiv Project

• [arXiv] [Distill] CoPhy: Distill to Think, Foresee to Act — Cognitive-Physical Reinforcement Learning for Autonomous Driving, arXiv 2026. arXiv

• [CVPR] [Preference] Drive My Way: Preference Alignment of Vision-Language-Action Model for Personalized Driving, CVPR 2026. arXiv Code

• [arXiv] [Preference] Causal Scene Narration with Runtime Safety Supervision for Vision-Language-Action Driving (CSN), arXiv 2026. arXiv

• [arXiv] [Preference] CPO++: Towards Robust Endogenous Reasoning — Unifying Drift Adaptation in Non-Stationary Tuning, arXiv 2026. arXiv

• [ICRA] [Preference] VL-DPO: Vision-Language-Guided Finetuning for Preference-Aligned Autonomous Driving, ICRA 2026. arXiv

• [arXiv] [RL] ThinkDrive: Chain-of-Thought Guided Progressive Reinforcement Learning Fine-Tuning for Autonomous Driving, arXiv 2026. arXiv

• [arXiv] [RL] FLARE: Learning Future-Aware Latent Representations from Vision-Language Models for Autonomous Driving, arXiv 2026. arXiv

• [CVPR] [RL] NoRD: A Data-Efficient Vision-Language-Action Model that Drives without Reasoning, CVPR 2026. arXiv Code

• [CVPR] [RL] MindDriver: Introducing Progressive Multimodal Reasoning for Autonomous Driving, CVPR 2026. arXiv Code

• [CVPR] [RL] PanoEnv: Exploring 3D Spatial Intelligence in Panoramic Environments with Reinforcement Learning, CVPR 2026. arXiv Code

• [CVPR] [RL] ELF-VLA: Unleashing VLA Potentials in Autonomous Driving via Explicit Learning from Failures, CVPR 2026. arXiv Code

• [arXiv] [RL] TakeVLA: Learning from Mistakes — Post-Training for Driving VLA with Takeover Data, arXiv 2026. arXiv

• [CVPR] [RL] Neuro-Cognitive Reward Modeling for Human-Centered Autonomous Vehicle Control, CVPR 2026. arXiv Project

• [arXiv] [RL] DreamerAD: Efficient Reinforcement Learning via Latent World Model for Autonomous Driving, arXiv 2026. arXiv

• [ICLR] [RL] AutoDrive-P³: Unified Chain of Perception-Prediction-Planning Thought via Reinforcement Fine-Tuning, ICLR 2026. arXiv Code

• [ECCV] [RL] ExploreVLA: Dense World Modeling and Exploration for End-to-End Autonomous Driving, ECCV 2026. arXiv Project

• [arXiv] [RL] SCORP: Scene-Consistent Multi-agent Diffusion Planning with Stable Online Reinforcement Post-Training for Cooperative Driving, arXiv 2026. arXiv

• [arXiv] [RL] FeaXDrive: Feasibility-aware Trajectory-Centric Diffusion Planning for End-to-End Autonomous Driving, arXiv 2026. arXiv

• [arXiv] [RL] SpanVLA: Efficient Action Bridging and Learning from Negative-Recovery Samples for Vision-Language-Action Model, arXiv 2026. arXiv Code

• [arXiv] [RL] DIAL: Driving Intents Amplify Planning-Oriented Reinforcement Learning, arXiv 2026. arXiv

• [arXiv] [RL] MAPLE: Latent Multi-Agent Play for End-to-End Autonomous Driving, arXiv 2026. arXiv

• [arXiv] [RL] SafeAlign-VLA: A Negative-Enhanced Safe Alignment Framework for Risk-Aware Autonomous Driving, arXiv 2026. arXiv

• [IJCNN] [RL] SARAD: LLM-Based Safety-Aware Hybrid Reinforcement Learning with Collision Prediction for Autonomous Driving, IJCNN 2026. arXiv

• [arXiv] [RL] BPF: Before Parc Fermé — RL-Time Pruning for Efficient Embodied LLMs in Autonomous Driving, arXiv 2026. arXiv

• [arXiv] [RL] DriveAnchor: Progressive Anchor-based Flow Learning for Autonomous Driving Planning, arXiv 2026. arXiv

• [CVPR] [Planner-RL] PlannerRFT: Reinforcing Diffusion Planners through Closed-Loop and Sample-Efficient Fine-Tuning, CVPR 2026. arXiv

• [arXiv] [Planner-RL] HDP: Unleashing the Potential of Diffusion Models for End-to-End Autonomous Driving, arXiv 2026. arXiv Code

• [arXiv] [Planner-RL] RAD-2: Scaling Reinforcement Learning in a Generator-Discriminator Framework, arXiv 2026. arXiv Project

• [arXiv] [SFT] LaST-VLA: Thinking in Latent Spatio-Temporal Space for Vision-Language-Action in Autonomous Driving, arXiv 2026. arXiv Code

• [arXiv] [SFT] VLM-AutoDrive: Post-Training Vision-Language Models for Safety-Critical Autonomous Driving Events, arXiv 2026. arXiv

• [CVPR] [SFT] The Blind Spot of Adaptation: Quantifying and Mitigating Forgetting in Fine-tuned Driving Models, CVPR 2026. arXiv

• [CVPR] [SFT] Learning Vision-Language-Action World Models for Autonomous Driving (VLA-World), CVPR 2026 (Findings). arXiv Project

• [arXiv] [SFT] EponaV2: Driving World Model with Comprehensive Future Reasoning, arXiv 2026. arXiv Code

• [arXiv] [Test-time] C-CoT: Counterfactual Chain-of-Thought with Vision-Language Models for Safe Autonomous Driving, arXiv 2026. arXiv

• [arXiv] [Test-time] Fast-dDrive: Efficient Block-Diffusion VLM for Autonomous Driving, arXiv 2026. arXiv

2025

• [ICLR] [Preference] TrajHF: Learning Personalized Driving Styles via Reinforcement Learning from Human Feedback, ICLR 2026. arXiv

• [NeurIPS] [Preference] DriveDPO: Policy Learning via Safety DPO for End-to-End Autonomous Driving, NeurIPS 2025. arXiv

• [RA-L] [Preference] TakeAD: Preference-Based Post-Optimization for End-to-End Autonomous Driving with Expert Takeover Data, IEEE RA-L 2025. arXiv

• [arXiv] [RL] AlphaDrive: Unleashing the Power of VLMs in Autonomous Driving via Reinforcement Learning and Reasoning, arXiv 2025. arXiv Code

• [arXiv] [RL] DriveMind: A Dual Visual Language Model-based Reinforcement Learning Framework for Autonomous Driving, arXiv 2025. arXiv

• [arXiv] [RL] Poutine: Vision-Language-Trajectory Pre-Training and Reinforcement Learning Post-Training Enable Robust End-to-End Autonomous Driving, arXiv 2025 (1st place, Waymo WOD-E2E Challenge). arXiv

• [NeurIPS] [RL] AutoVLA: A Vision-Language-Action Model for End-to-End Autonomous Driving with Adaptive Reasoning and Reinforcement Fine-Tuning, NeurIPS 2025. arXiv Code

• [AAAI] [RL] Drive-R1: Bridging Reasoning and Planning in VLMs for Autonomous Driving with Reinforcement Learning, AAAI 2026. arXiv Code

• [ICCV-W] [RL] LaViPlan: Language-Guided Visual Path Planning with RLVR, ICCV 2025 Workshop. arXiv

• [ICLR] [RL] DriveAgent-R1: Advancing VLM-based Autonomous Driving with Active Perception and Hybrid Thinking, ICLR 2026. arXiv Code

• [arXiv] [RL] IRL-VLA: Training a Vision-Language-Action Policy via Reward World Model, arXiv 2025. arXiv Code

• [ICLR] [RL] AutoDrive-R²: Incentivizing Reasoning and Self-Reflection Capacity for VLA Model in Autonomous Driving, ICLR 2026. arXiv Code

• [ICRA] [RL] AdaThinkDrive: Adaptive Thinking via Reinforcement Learning for Autonomous Driving, ICRA 2026. arXiv Code

• [arXiv] [RL] VDRive: Leveraging Reinforced VLA and Diffusion Policy for End-to-End Autonomous Driving, arXiv 2025. arXiv

• [arXiv] [RL] Alpamayo-R1: Bridging Reasoning and Action Prediction for Generalizable Autonomous Driving in the Long Tail, arXiv 2025 (NVIDIA). arXiv Code

• [arXiv] [RL] WAM-Diff: A Masked Diffusion VLA Framework with MoE and Online Reinforcement Learning for Autonomous Driving, arXiv 2025. arXiv Code

• [CVPR] [RL] OmniDrive-R1: Reinforcement-driven Interleaved Multi-modal Chain-of-Thought for Trustworthy Vision-Language Autonomous Driving, CVPR 2026. arXiv

• [NeurIPS] [Planner-RL] RAD: Training an End-to-End Driving Policy via Large-Scale 3DGS-based Reinforcement Learning, NeurIPS 2025. arXiv Code

• [ICLR] [Planner-RL] Plan-R1: Safe and Feasible Trajectory Planning as Language Modeling, ICLR 2026. arXiv Code

• [arXiv] [Planner-RL] EvaDrive: Evolutionary Adversarial Policy Optimization for End-to-End Autonomous Driving, arXiv 2025. arXiv

• [arXiv] [Planner-RL] TSA-MPR: Autoregressive End-to-End Planning with Time-Invariant Spatial Alignment and Multi-Objective Policy Refinement, arXiv 2025. arXiv Project

• [arXiv] [Planner-RL] DiffusionDriveV2: Reinforcement Learning-Constrained Truncated Diffusion Modeling in End-to-End Autonomous Driving, arXiv 2025. arXiv Code

• [arXiv] [SFT] CoReVLA: A Dual-Stage End-to-End Autonomous Driving Framework for Long-Tail Scenarios via Collect-and-Refine, arXiv 2025. arXiv Code

• [arXiv] [SFT] Reasoning-VLA: A Fast and General Vision-Language-Action Reasoning Model for Autonomous Driving, arXiv 2025. arXiv Code

• [arXiv] [SFT] LLaViDA: A Large Language Vision Driving Assistant for Explicit Reasoning and Enhanced Trajectory Planning, arXiv 2025. arXiv

• [arXiv] [Test-time] ReflectDrive: Discrete Diffusion for Reflective Vision-Language-Action Models in Autonomous Driving, arXiv 2025. arXiv Code

• [ICRA] [Test-time] DriveCritic: Towards Context-Aware, Human-Aligned Evaluation for Autonomous Driving with Vision-Language Models, ICRA 2026. arXiv Project

• [arXiv] [Test-time] Post-Training and Test-Time Scaling of Generative Agent Behavior Models for Interactive Autonomous Driving, arXiv 2025. arXiv

• [CVPR] [Test-time] Counterfactual VLA: Self-reflective Vision-Language-Action Model with Adaptive Reasoning, CVPR 2026. arXiv

2024

• [CoRL] [Distill] VLM-AD: End-to-End Autonomous Driving through Vision-Language Model Supervision, CoRL 2025. arXiv

• [arXiv] [Distill] RAPID: Robust RL with LLM-Driven Data Synthesis and Policy Adaptation for Autonomous Driving, arXiv 2024. arXiv

• [UbiComp] [RL] Optimizing Autonomous Driving for Safety: A Human-Centric Approach with LLM-Enhanced RLHF, UbiComp/ISWC 2024 Companion (ACM). arXiv

---

📊 Benchmarks & Simulators

Post-training targets closed-loop driving quality, so evaluation increasingly relies on (pseudo-)closed-loop simulators and driving-specific scores.

• NAVSIM: Data-Driven Non-Reactive Autonomous Vehicle Simulation and Benchmarking, NeurIPS 2024 (Datasets & Benchmarks). One-shot non-reactive simulation with the PDMS score. arXiv Code

• NAVSIM v2, CoRL 2025. Two-stage pseudo-closed-loop protocol with the EPDMS score and the navhard split. arXiv Code

• Bench2Drive: Towards Multi-Ability Benchmarking of Closed-Loop End-To-End Autonomous Driving, NeurIPS 2024 (Datasets & Benchmarks). Fully closed-loop on CARLA with reactive traffic; reports Driving Score (DS) and Success Rate (SR). arXiv Code

• WOD-E2E: Waymo Open Dataset for End-to-End Driving in Challenging Long-Tail Scenarios, CVPR 2026. Long-tail open-loop benchmark with the Rater Feedback Score (RFS). arXiv Code

• nuScenes: A Multimodal Dataset for Autonomous Driving, CVPR 2020. Standard open-loop benchmark (L2 / collision rate). arXiv Code

• nuReasoning: A Reasoning-Centric Dataset and Benchmark for Long-Tail Autonomous Driving, arXiv 2026. NAVSIM-style safety-gated NPS score with 5-second ADE. arXiv Project

• AlpaSim: A Modular, Lightweight, and Data-Driven Research Simulator for Autonomous Driving, NVIDIA 2025. Neural-rendering photorealistic sensor simulation for closed-loop testing. Code

• NVIDIA OmniDreams: Real-Time Generative World Model for Closed-Loop Autonomous Vehicle Simulation, arXiv 2026. arXiv Code

---

Related Surveys

• Vision-Language-Action Models for Autonomous Driving: Past, Present, and Future, arXiv 2025. arXiv Repo

• A Survey on Vision-Language-Action Models for Autonomous Driving, ICCV 2025 Workshop. arXiv Repo

• A Survey for Foundation Models in Autonomous Driving, ICCVDM 2025. arXiv

• Survey of General End-to-End Autonomous Driving: A Unified Perspective, 2025. TechRxiv

• A Survey on End-to-End Autonomous Driving Training from the Perspectives of Data, Strategy, and Platform, IEEE T-ITS 2026. Repo

---

Citation

If you find this repository or our survey useful, please consider citing:

@article{yang2026post,
  title={Post-Training in End-to-End Autonomous Driving},
  author={Yang, Ruining and Wang, Muxing and Chen, Yixiao and Guo, Tongfei and Xu, Yi and Cui, Can and Yang, Zichong and Zhang, Yitian and Wang, Ziran and Fu, Yun and others},
  journal={arXiv preprint arXiv:2607.08072},
  year={2026}
}

---

🤝 Contributing

Contributions are welcome!

You can simply open an issue to suggest a paper, report a broken link, or correct a venue.

---

⭐ Star History

If this list helps your research, please consider giving it a star. It helps others discover it.