Research map / Vision-language and multimodal / Robot action and navigation
Discrete Diffusion for Reflective Vision-Language-Action Models in Autonomous Driving
Pengxiang Li, Yinan Zheng, Yue Wang, Huimin Wang, Hang Zhao, Jingjing Liu, Xianyuan Zhan, Kun Zhan, XianPeng Lang
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 reflective inference (goal-conditioned generation and safety-guided regeneration) (ablation) | ReflectDrive · NAVSIM | PDMS: 84.8 -> 91.1 | Improved |
| add goal-conditioned generation only (ablation) | ReflectDrive · NAVSIM | EP: 79.0 -> 83.8 | Improved |
| add goal-conditioned generation only (ablation) | ReflectDrive · NAVSIM | TTC: 92.2 -> 91.5 | Hurt |
| add safety-guided regeneration only (ablation) | ReflectDrive · NAVSIM | PDMS: 84.8 -> 90.3 | Improved |
| add goal-conditioned generation on top of safety-guided regeneration (ablation) | ReflectDrive · NAVSIM | PDMS: 90.3 -> 91.1 | Improved |
| use ground-truth agent states instead of constant-velocity agents in the reflection reward (ablation) | ReflectDrive · NAVSIM | PDMS: 91.1 -> 94.7 | Improved |
| increase diffusion generation steps beyond 5 (ablation) | ReflectDrive (w/o R.I.) · NAVSIM | PDMS | Hurt |
| increase number of goal points (multi-modal candidates) (ablation) | ReflectDrive · NAVSIM | PDMS | Improved |
| increase exploration steps and max regeneration iterations (ablation) | ReflectDrive · NAVSIM | PDMS | Improved |
| more inference opportunities for reflection | ReflectDrive · NAVSIM | PDMS | No clear effect |
Closest papers by meaning
- Intend, Reflect, Refine: An Adaptive Multimodal Reflection Framework for Autonomous Driving (NeurIPS 2026 · similarity 0.90)
- ReflectDrive-2: Reinforcement-Learning-Aligned Self-Editing for Discrete Diffusion Driving (NeurIPS 2026 · similarity 0.90)
- DiffusionDrive: Truncated Diffusion Model for End-to-End Autonomous Driving (CVPR 2025 · similarity 0.89)
- AutoVLA: A Vision-Language-Action Model for End-to-End Autonomous Driving with Adaptive Reasoning and Reinforcement Fine-Tuning (NeurIPS 2025 · similarity 0.89)
- Diffusion-Based Planning for Autonomous Driving with Flexible Guidance (ICLR 2025 · similarity 0.89)
- Diffusion-ES: Gradient-free Planning with Diffusion for Autonomous and Instruction-guided Driving (CVPR 2024 · 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.
