Research map / Vision-language and multimodal / Robot action and navigation
AutoDrive-R²: Incentivizing Reasoning and Self-Reflection Capacity for VLA Model in Autonomous Driving
Zhenlong Yuan, Chengxuan Qian, Jing Tang, Rui Chen, Zijian Song, Lei Sun, Xiangxiang Chu, Yujun Cai, Dapeng Zhang, Shuo Li
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add SFT on nuScenesR2-6K CoT data (ablation) | Qwen2.5-VL-7B · nuScenes (L2 error) | avg L2 error (m): 1.45 -> 0.27 | Improved |
| add RL stage after SFT (two-stage vs RL only) (ablation) | AutoDrive-R² 7B · nuScenes (L2 error) | avg L2 error (m): RL-only worse than SFT by 22.2%; SFT+RL 0.19 | Improved |
| add four-step reasoning structure to CoT data (ablation) | AutoDrive-R² 7B · nuScenes (L2 error) | avg L2 error (m): 0.25 -> 0.19 | Improved |
| add self-reflection step to CoT data (ablation) | AutoDrive-R² 7B · nuScenes (L2 error) | avg L2 error (m): 0.23 -> 0.19 | Improved |
| add spatial alignment reward r_pos (ablation) | AutoDrive-R² 7B · nuScenes (L2 error) | avg L2 error (m): 0.53 -> 0.19 | Improved |
| add steering reward r_ste (ablation) | AutoDrive-R² 7B · nuScenes (L2 error) | avg L2 error (m): 0.21 -> 0.19 | Improved |
| add velocity reward r_vel (ablation) | AutoDrive-R² 7B · nuScenes (L2 error) | avg L2 error (m): 0.22 -> 0.19 | Improved |
| add temporal smoothness reward r_tem (ablation) | AutoDrive-R² 7B · nuScenes (L2 error) | avg L2 error (m): 0.24 -> 0.19 | Improved |
| use uniform instead of decaying reward weights (ablation) | AutoDrive-R² 7B · nuScenes (L2 error) | avg L2 error (m): 0.22 -> 0.19 | Improved |
| set KL coefficient beta to 0.04 vs 0.02 or 0.06 (ablation) | AutoDrive-R² 7B · nuScenes (L2 error) | avg L2 error (m): 0.21 (0.02), 0.20 (0.06) -> 0.19 (0.04) | Improved |
| increase GRPO generations G from 2 to 6 (ablation) | AutoDrive-R² 7B · nuScenes (L2 error) | avg L2 error (m): 0.23 -> 0.19 | Improved |
| increase GRPO generations G from 6 to 8 (ablation) | AutoDrive-R² 7B · nuScenes (L2 error) | avg L2 error (m) | No clear effect |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- EMMA+ — EMMA: Empowering Multi-modal Mamba with Structural and Hierarchical Alignment (ICLR 2025) · better on 2 of 2 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- AutoVLA: A Vision-Language-Action Model for End-to-End Autonomous Driving with Adaptive Reasoning and Reinforcement Fine-Tuning (NeurIPS 2025 · similarity 0.94)
- Drive-R1: Bridging Reasoning and Planning in VLMs for Autonomous Driving with Reinforcement Learning (AAAI 2026 · similarity 0.93)
- LaST-VLA: Thinking in Latent Spatio-Temporal Space for Vision-Language-Action in Autonomous Driving (NeurIPS 2026 · similarity 0.92)
- $AutoDrive\text{-}P^3$: Unified Chain of Perception–Prediction–Planning Thought via Reinforcement Fine-Tuning (ICLR 2026 · similarity 0.92)
- DriveAgent-R1: Advancing VLM-based Autonomous Driving with Active Perception and Hybrid Thinking (ICLR 2026 · similarity 0.91)
- Distill to Think, Foresee to Act: Cognitive-Physical Reinforcement Learning for Autonomous Driving (NeurIPS 2026 · similarity 0.91)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
