Research map / Vision-language and multimodal / Robot action and navigation
Latent Chain-of-Thought World Modeling for End-to-End Autonomous Driving
Shuhan Tan, Kashyap Chitta, Yuxiao Chen, Ran Tian, Yurong You, Yan Wang, Wenjie Luo, Yulong Cao, Philipp Krähenbühl, Marco Pavone, Boris Ivanovic
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 |
|---|---|---|---|
| replace no reasoning with latent CoT + RL | Qwen3-0.5B VLA · PhysicalAI-AV | ADE: 1.762 -> 1.626 | Improved |
| latent CoT vs text CoT before RL | LCDrive · PhysicalAI-AV | ADE: 1.650 -> 1.668 | No clear effect |
| add RL post-training to latent CoT (predicted LWM) (ablation) | LCDrive · PhysicalAI-AV | ADE: 1.668 -> 1.626 | Improved |
| add RL to latent CoT (GT LWM) (ablation) | Latent CoT* · PhysicalAI-AV | ADE: 1.268 -> 1.197 | Improved |
| add RL to history-only LWM (no rollout) (ablation) | LWM0-only* · PhysicalAI-AV | ADE: 1.393 -> 1.397 | No clear effect |
| add RL to history-only LWM (ablation) | LWM0-only* · PhysicalAI-AV | OffRoad5.0: 3.104 -> 4.218 | Hurt |
| add interleaved latent CoT rollout over history-only LWM (GT) (ablation) | LWM0-only* · PhysicalAI-AV | ADE: 1.393 -> 1.268 | Improved |
| latent CoT instead of text CoT after RL | Text CoT · PhysicalAI-AV | ADE: 1.650 -> 1.626 | Improved |
| latent CoT vs text CoT per scenario | Text CoT · Cut-In and Lead Vehicle Following | ADE: 1.884 -> 2.385 | Hurt |
| latent CoT vs text CoT per scenario | Text CoT · Speed Control | ADE: 2.037 -> 1.376 | Improved |
| latent CoT instead of text CoT | Text CoT · reasoning latency | speedup: 1.8x | Improved |
Closest papers by meaning
- OneVL: One-Step Latent Reasoning and Planning with Vision-Language Explanation (NeurIPS 2026 · similarity 0.94)
- LaST-VLA: Thinking in Latent Spatio-Temporal Space for Vision-Language-Action in Autonomous Driving (NeurIPS 2026 · similarity 0.94)
- Drive-R1: Bridging Reasoning and Planning in VLMs for Autonomous Driving with Reinforcement Learning (AAAI 2026 · similarity 0.92)
- AutoVLA: A Vision-Language-Action Model for End-to-End Autonomous Driving with Adaptive Reasoning and Reinforcement Fine-Tuning (NeurIPS 2025 · similarity 0.92)
- $AutoDrive\text{-}P^3$: Unified Chain of Perception–Prediction–Planning Thought via Reinforcement Fine-Tuning (ICLR 2026 · similarity 0.91)
- NoRD: A Data-Efficient Vision-Language-Action Model that Drives without Reasoning (CVPR 2026 · similarity 0.90)
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.
