Research map / Reinforcement learning / Visual and representation RL
Unleashing VLA Potentials in Autonomous Driving via Explicit Learning from Failures
Yuechen Luo, Fang Li, Qimao Chen, Shaoqing Xu, Jiaxin Liu, Ziying Song, Zhi-xin Yang, Fuxi Wen
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 teacher-feedback-guided refinement samples to GRPO | GRPO · NAVSIMv1 | PDMS: 89.0 -> 91.0 | Improved |
| add ground-truth trajectories directly to GRPO batch (ablation) | GRPO · NAVSIMv1 | PDMS: 89.0 -> 89.2 | Improved |
| replace rule-based feedback with teacher structured feedback (ablation) | Rule-GRPO · NAVSIMv1 | PDMS: 89.6 -> 91.0 | Improved |
| replace GT trajectory injection with feedback-guided refinement (ablation) | GT-GRPO · NAVSIMv1 | PDMS: 89.2 -> 91.0 | Improved |
| add teacher feedback refinement | GRPO · NAVSIMv1 high-level planning | accuracy (%): 79.3 -> 80.3 | Improved |
| add teacher feedback refinement | GRPO · RL total-failure ratio | PDMS total-failure rate (%): 2.73 -> 1.08 | Improved |
| curate difficult and ambiguous 24k samples instead of full 85k (ablation) | ELF-VLA · NAVSIMv1 | PDMS: 89.1 -> 91.0 | Improved |
| curate 24k instead of random 24k (ablation) | ELF-VLA · NAVSIMv1 | PDMS: 88.9 -> 91.0 | Improved |
| use k=1 refinement response instead of k=4 (ablation) | ELF-VLA · NAVSIMv1 | PDMS: 89.0 -> 91.0 | Improved |
| use k=1 instead of k=2 refinement responses (ablation) | ELF-VLA · NAVSIMv1 | PDMS: 89.7 -> 91.0 | Improved |
| add policy shaping on feedback samples (ablation) | ELF-VLA · NAVSIMv1 | PDMS: 89.3 -> 91.0 | Improved |
| use ELF-VLA vs DriveSuprim | DriveSuprim · NAVSIMv2 | EPDMS: 87.1 -> 87.1 | No clear effect |
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 0.92)
- Drive-R1: Bridging Reasoning and Planning in VLMs for Autonomous Driving with Reinforcement Learning (AAAI 2026 · similarity 0.91)
- NoRD: A Data-Efficient Vision-Language-Action Model that Drives without Reasoning (CVPR 2026 · similarity 0.91)
- Impromptu VLA: Open Weights and Open Data for Driving Vision-Language-Action Models (NeurIPS 2025 · similarity 0.91)
- LaST-VLA: Thinking in Latent Spatio-Temporal Space for Vision-Language-Action in Autonomous Driving (NeurIPS 2026 · similarity 0.90)
- FIVE-VLA: Fast and EffectIVE Closed-Loop Autonomous Driving with Recurrent Action Memory (NeurIPS 2026 · similarity 0.89)
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.
