mlsanityData to 14 Oct 2026

AdaThinkDrive: Adaptive Thinking via Reinforcement Learning for Autonomous Driving

arXiv 2509.13769 ↗First posted 17 Sep 2025cs.CVcited by 58

On the map: LLM reasoning → Chain-of-thought reasoning for driving

also in 2 more
Abstract (arXiv)

While reasoning technology like Chain of Thought (CoT) has been widely adopted in Vision Language Action (VLA) models, it demonstrates promising capabilities in end to end autonomous driving. However, recent efforts to integrate CoT reasoning often fall short in simple scenarios, introducing unnecessary computational overhead without improving decision quality. To address this, we propose AdaThinkDrive, a novel VLA framework with a dual mode reasoning mechanism inspired by fast and slow thinking. First, our framework is pretrained on large scale autonomous driving (AD) scenarios using both question answering (QA) and trajectory datasets to acquire world knowledge and driving commonsense. During supervised fine tuning (SFT), we introduce a two mode dataset, fast answering (w/o CoT) and slow thinking (with CoT), enabling the model to distinguish between scenarios that require reasoning. Furthermore, an Adaptive Think Reward strategy is proposed in conjunction with the Group Relative Policy Optimization (GRPO), which rewards the model for selectively applying CoT by comparing trajectory quality across different reasoning modes. Extensive experiments on the Navsim benchmark show that AdaThinkDrive achieves a PDMS of 90.3, surpassing the best vision only baseline by 1.7 points. Moreover, ablations show that AdaThinkDrive surpasses both the never Think and always Think baselines, improving PDMS by 2.0 and 1.4, respectively. It also reduces inference time by 14% compared to the always Think baseline, demonstrating its ability to balance accuracy and efficiency through adaptive reasoning.

Key facts. Cited by 58 papers in the map (data through 14 Oct 2026). Placed in Chain-of-thought reasoning for driving, one of 17 directions in LLM reasoning. Builds on 38 of its 41 references that are in the map. First posted on arXiv 17 Sep 2025.

Builds on

38 of 41 references.

Closest ancestors · 5

Same direction, newest first

  1. 2025Chain-of-Thought for Autonomous Driving: A Comprehensive Survey and Future Prospects26 May
  2. 2025ReasonPlan: Unified Scene Prediction and Decision Reasoning for Closed-loop Autonomous Driving26 May
  3. 2025OmniDrive: A Holistic Vision-Language Dataset for Autonomous Driving with Counterfactual Reasoning6 Apr
  4. 2025ORION: A Holistic End-to-End Autonomous Driving Framework by Vision-Language Instructed Action Generation25 Mar
  5. 2025AlphaDrive: Unleashing the Power of VLMs in Autonomous Driving via Reinforcement Learning and Reasoning10 Mar

Other references · 33

  1. 2024DeepSeekMath: Pushing the Limits of Mathematical Reasoning in Open Language Modelscited by 9,379
  2. 2025InternVL3: Exploring Advanced Training and Test-Time Recipes for Open-Source Multimodal Modelscited by 1,834
  3. 2022Planning-oriented Autonomous Drivingcited by 1,087
  4. 2023Vision-Language Models for Vision Tasks: A Surveycited by 839
  5. 2023VAD: Vectorized Scene Representation for Efficient Autonomous Drivingcited by 622
  6. 2023DriveLM: Driving with Graph Visual Question Answeringcited by 536
  7. 2023End-to-end Autonomous Driving: Challenges and Frontierscited by 490
  8. 2024DriveVLM: The Convergence of Autonomous Driving and Large Vision-Language Modelscited by 490
  9. 2022TransFuser: Imitation with Transformer-Based Sensor Fusion for Autonomous Drivingcited by 485
  10. 2024NAVSIM: Data-Driven Non-Reactive Autonomous Vehicle Simulation and Benchmarkingcited by 354
  11. 2024DiffusionDrive: Truncated Diffusion Model for End-to-End Autonomous Drivingcited by 334
  12. 2023NuScenes-QA: A Multi-modal Visual Question Answering Benchmark for Autonomous Driving Scenariocited by 262
21 more
  1. 2024Hydra-MDP: End-to-end Multimodal Planning with Multi-target Hydra-Distillationcited by 249
  2. 2024EMMA: End-to-End Multimodal Model for Autonomous Drivingcited by 231
  3. 2025ReCogDrive: A Reinforced Cognitive Framework for End-to-End Autonomous Drivingcited by 198
  4. 2025ThinkPrune: Pruning Long Chain-of-Thought of LLMs via Reinforcement Learningcited by 192
  5. 2024Senna: Bridging Large Vision-Language Models and End-to-End Autonomous Drivingcited by 156
  6. 2025End-to-End Driving with Online Trajectory Evaluation via BEV World Modelcited by 151
  7. 2025AdaptThink: Reasoning Models Can Learn When to Thinkcited by 147
  8. 2023LingoQA: Visual Question Answering for Autonomous Drivingcited by 145
  9. 2025GoalFlow: Goal-Driven Flow Matching for Multimodal Trajectories Generation in End-to-End Autonomous Drivingcited by 120
  10. 2025Thinkless: LLM Learns When to Thinkcited by 113
  11. 2024OpenEMMA: Open-Source Multimodal Model for End-to-End Autonomous Drivingcited by 93
  12. 2024DRAMA: An Efficient End-to-end Motion Planner for Autonomous Driving with Mambacited by 87
  13. 2025AdaCoT: Pareto-Optimal Adaptive Chain-of-Thought Triggering via Reinforcement Learningcited by 87
  14. 2025Concise Reasoning via Reinforcement Learningcited by 79
  15. 2024Holistic Autonomous Driving Understanding by Bird's-Eye-View Injected Multi-Modal Large Modelscited by 76
  16. 2025ShorterBetter: Guiding Reasoning Models to Find Optimal Inference Length for Efficient Reasoningcited by 72
  17. 2025A Survey on Vision-Language-Action Models for Autonomous Drivingcited by 72
  18. 2025Learning When to Think: Shaping Adaptive Reasoning in R1-Style Models via Multi-Stage RLcited by 62
  19. 2025Impromptu VLA: Open Weights and Open Data for Driving Vision-Language-Action Modelscited by 60
  20. 2025Hydra-NeXt: Robust Closed-Loop Driving with Open-Loop Trainingcited by 39
  21. 2025Sce2DriveX: A Generalized MLLM Framework for Scene-to-Drive Learningcited by 25

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2025: 6; 2026: 52.

58 papers.

Sort:

By area:

  1. 2025Alpamayo-R1: Bridging Reasoning and Action Prediction for Generalizable Autonomous Driving in the Long Tailcited by 146
  2. 2025A Survey on Efficient Vision-Language-Action Modelscited by 34
  3. 2025Vision-Language-Action Models for Autonomous Driving: Past, Present, and Futurecited by 33
  4. 2026LaST-VLA: Thinking in Latent Spatio-Temporal Space for Vision-Language-Action in Autonomous Drivingcited by 28
  5. 2025MindDrive: A Vision-Language-Action Model for Autonomous Driving via Online Reinforcement Learningcited by 25
  6. 2026DriveVA: Video Action Models are Zero-Shot Driverscited by 23
  7. 2025Counterfactual VLA: Self-Reflective Vision-Language-Action Model with Adaptive Reasoningcited by 22
  8. 2026VGGDrive: Empowering Vision-Language Models with Cross-View Geometric Grounding for Autonomous Drivingcited by 20
  9. 2026DriveFine: Refining-Augmented Masked Diffusion VLA for Precise and Robust Drivingcited by 19
  10. 2026Unleashing VLA Potentials in Autonomous Driving via Explicit Learning from Failurescited by 19
  11. 2026UniDriveVLA: Unifying Understanding, Perception, and Action Planning for Autonomous Drivingcited by 19
  12. 2026NoRD: A Data-Efficient Vision-Language-Action Model that Drives without Reasoningcited by 15
38 more
  1. 2026Devil is in Narrow Policy: Unleashing Exploration in Driving VLA Modelscited by 15
  2. 2025dVLM-AD: Enhance Diffusion Vision-Language-Model for Driving via Controllable Reasoningcited by 14
  3. 2026DVGT-2: Vision-Geometry-Action Model for Autonomous Driving at Scalecited by 13
  4. 2026CoWorld-VLA: Thinking in a Multi-Expert World Model for Autonomous Drivingcited by 12
  5. 2026DynVLA: Learning World Dynamics for Action Reasoning in Autonomous Drivingcited by 11
  6. 2026Metis: A Generalizable and Efficient World-Action Model for Autonomous Driving and Urban Navigationcited by 11
  7. 2026MindDriver: Introducing Progressive Multimodal Reasoning for Autonomous Drivingcited by 10
  8. 2026Bench2Drive-VL: Benchmarks for Closed-Loop Autonomous Driving with Vision-Language Modelscited by 10
  9. 2026Xiaomi OneVL: One-Step Latent Reasoning and Planning with Vision-Language Explanationcited by 10
  10. 2026MindVLA-U1: VLA Beats VA with Unified Streaming Architecture for Autonomous Drivingcited by 6
  11. 2026Auto-JEPA: A Latent World Model of Continuous Intent for End-to-End Autonomous Drivingcited by 5
  12. 2026Distill to Think, Foresee to Act: Cognitive-Physical Reinforcement Learning for Autonomous Drivingcited by 3
  13. 2026VLGA: Vision-Language-Geometry-Action Models for Autonomous Drivingcited by 3
  14. 2026IDOL: Inverse-Dynamics-Guided Future Prediction for End-to-End Autonomous Drivingcited by 2
  15. 2026DriveReward: A Comprehensive Dataset and Generative Vision-Language Reward Model for Autonomous Drivingcited by 2
  16. 2026Intend, Reflect, Refine: An Adaptive Multimodal Reflection Framework for Autonomous Drivingcited by 2
  17. 2026Post-Training in End-to-End Autonomous Drivingcited by 2
  18. 2026GeoWorldAD: Geometry World Action Model for Autonomous Drivingcited by 2
  19. 2026UniTeD: Unified Temporal Diffusion for Joint Perception and Planning in Autonomous Drivingcited by 1
  20. 2026DriveVer: Lightweight Trajectory Evaluator as Test-Time Verifier for Autonomous Drivingcited by 1
  21. 2026Teaching Vision-Language-Action Models What to See and Where to Lookcited by 1
  22. 2026Do Vision-Language-Action Models Mean What They Say? On the Role of Faithfulness in Embodied Reasoningcited by 1
  23. 2026WCog-VLA: A Dual-Level World-Cognitive Vision-Language-Action Model for End-to-End Autonomous Drivingcited by 1
  24. 2026Large Discrete Policy: Advancing Explicit Behavior Modeling with Stochastic Iterative Scoringcited by 1
  25. 2026Open-Ended Instruction Realization with LLM-Enabled Multi-Planner Scheduling in Autonomous Vehiclescited by 0
  26. 2026Latency Analysis and Optimization of Alpamayo 1 via Efficient Trajectory Generationcited by 0
  27. 2026GeoDrive-Bench: Benchmarking Region-Specific Multimodal Reasoning in Autonomous Drivingcited by 0
  28. 2026LiAuto-GeoX: Efficient Grounded Driving Transformercited by 0
  29. 2026VeriDrive: Verifiable Counterfactual Supervision for Cost-Efficient Vision-Language Planningcited by 0
  30. 2026WorkDrive: Roadwork Chain of Causation for Autonomous Drivingcited by 0
  31. 2026Cognitive Dual-Process Planning for Autonomous Driving with Structured Scene Knowledge and Verifiable Reasoning-Action Consistencycited by 0
  32. 2026MOJITO: Modal Joint Learning for Unified End-to-End Autonomous Drivingcited by 0
  33. 2026One Policy, Many Embodiments: Unified Camera-Centric Action Geometry Pre-training for Heterogeneous Embodied Manipulationcited by 0
  34. 2026Drive the Thoughts: Runtime Monitoring of VLA Reasoning-Trajectory Consistencycited by 0
  35. 2026Aligning Multi-Trajectory Supervision with Policy Optimization for VLA Drivingcited by 0
  36. 2026Beyond Textual Chain-of-Thought: A Survey on Action-Grounded Reasoning in Autonomous Drivingcited by 0
  37. 2026Drive-HWM: Hierarchical World Models for Dynamic-Latent Guided Autonomous Drivingcited by 0
  38. 2026READ: Learning Risk-Informed Fields for End-to-End Autonomous Drivingcited by 0

50 most cited of 58 All 58 (JSON)

Similar, not cited

Close in wording, no citation either way

  1. 2026ThinkDrive: Chain-of-Thought Guided Progressive Reinforcement Learning Fine-Tuning for Autonomous Driving
  2. 2025AutoDrive-R2: Incentivizing Reasoning and Self-Reflection Capacity for VLA Model in Autonomous Driving
  3. 2025AutoVLA: A Vision-Language-Action Model for End-to-End Autonomous Driving with Adaptive Reasoning and Reinforcement Fine-Tuning
  4. 2026CritiqueDriveVLM: From Verifier-Guided Reinforcement Learning to Latent Thought Distillation for Autonomous Driving
  5. 2025OpenREAD: Reinforced Open-Ended Reasoning for End-to-End Autonomous Driving with LLM-as-Critic
  6. 2025Drive-R1: Bridging Reasoning and Planning in VLMs for Autonomous Driving with Reinforcement Learning
  7. 2025CoT4AD: A Vision-Language-Action Model with Explicit Chain-of-Thought Reasoning for Autonomous Driving
  8. 2025Latent Chain-of-Thought World Modeling for End-to-End Driving