Research map / Vision-language and multimodal / Robot action and navigation
Reasoning-VLA: An Efficient and Spatial-Guided General Vision-Language-Action Reasoning Model for Autonomous Driving
Dapeng Zhang, Zhenlong Yuan, Zhangquan Chen, Chih-Ting Liao, Yinda Chen, Fei Shen, Qingguo Zhou, Tat-Seng Chua
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 learnable action queries instead of non-learnable queries (ablation) | Reasoning-VLA-7B · nuScenes subset of unified dataset | avg L2 (m): 0.30 -> 0.23 | Improved |
| add implicit spatial guidance initialization of action queries (ablation) | Reasoning-VLA-7B · nuScenes subset of unified dataset | avg L2 (m): 0.27 -> 0.23 | Improved |
| add action refinement module (ablation) | Reasoning-VLA-7B · nuScenes subset of unified dataset | avg L2 (m): 0.29 -> 0.23 | Improved |
| add RL (GRPO) fine-tuning after SFT (ablation) | Reasoning-VLA-7B · nuScenes subset of unified dataset | avg L2 (m): 0.26 -> 0.23 | Improved |
| add RL after SFT when ARM is removed (ablation) | R-VLA(w/o ARM) · nuScenes subset of unified dataset | avg L2 (m): 0.29 -> 0.29 | No clear effect |
| add RL after SFT on zero-shot unseen datasets (ablation) | Reasoning-VLA-7B · Zero-shot unified dataset | avg L2 (m): 0.31 -> 0.29 | Improved |
| add RL to SFT on the unified dataset (per sub-dataset) (ablation) | Reasoning-VLA-7B · Unified dataset sub-datasets | avg L2 (m): 0.24 -> 0.23 | No clear effect |
| scale backbone from 3B to 7B | Reasoning-VLA · nuScenes open-loop | avg L2 (m): 0.30 -> 0.23 | Improved |
| additional RL on nuScenes training clips (7B+) | Reasoning-VLA-7B · nuScenes open-loop | avg L2 (m): 0.23 -> 0.22 | Improved |
| additional RL on nuScenes training clips (7B+) | Reasoning-VLA-7B · NeuroNCAP closed-loop | avg NeuroNCAP score: 2.25 -> 2.19 | Hurt |
| additional RL on nuScenes training clips (7B+) | Reasoning-VLA-7B · NeuroNCAP closed-loop | avg collision rate (%): 59.4 -> 59.8 | Hurt |
| use Reasoning-VLA-7B instead of Para-Drive on NAVSIM | NAVSIM planner · NAVSIM | PDMS: 84.0 -> 91.7 | Improved |
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 1 of 1 shared result
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 0.95)
- Impromptu VLA: Open Weights and Open Data for Driving Vision-Language-Action Models (NeurIPS 2025 · similarity 0.93)
- LaST-VLA: Thinking in Latent Spatio-Temporal Space for Vision-Language-Action in Autonomous Driving (NeurIPS 2026 · similarity 0.93)
- Drive-R1: Bridging Reasoning and Planning in VLMs for Autonomous Driving with Reinforcement Learning (AAAI 2026 · similarity 0.92)
- Less Language, More Latents: Annotation-Efficient VLAs for Driving (NeurIPS 2026 · similarity 0.91)
- NoRD: A Data-Efficient Vision-Language-Action Model that Drives without Reasoning (CVPR 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.
