Research map / Large language models / Agents, code and math
Making Large Language Models Better Planners with Reasoning-Decision Alignment
Zhijian Huang, Tao Tang, Shaoxiang Chen, Sihao Lin, Zequn Jie, Lin Ma, Guangrun Wang, Xiaodan Liang
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 reasoning-decision alignment loss (ablation) | RDA-Driver · DriveLM-nuScenes CoT score | CoT score: 62 -> 71 | Improved |
| add ranking and binary alignment losses together (ablation) | RDA-Driver · nuScenes val (UniAD metrics) | L2 avg (m): 0.85 -> 0.80 | Improved |
| add ranking and binary alignment losses together (ablation) | RDA-Driver · nuScenes val (UniAD metrics) | Collision avg (%): 0.44 -> 0.32 | Improved |
| add binary loss only (without ranking) (ablation) | RDA-Driver · nuScenes val (UniAD metrics) | L2 avg (m): 0.85 -> 0.87 | No clear effect |
| use BEV features flattened by region instead of CLIP encoder (ablation) | RDA-Driver · nuScenes val (UniAD metrics) | Collision avg (%): 0.48 -> 0.41 | Improved |
| replace BEV flattening with BEV pooling (ablation) | RDA-Driver · nuScenes val (UniAD metrics) | L2 avg (m): 0.84 -> 1.35 | Hurt |
| use multi-turn CoT instead of single-turn with all CoT types (ablation) | RDA-Driver · nuScenes val (UniAD metrics) | L2 avg (m): 0.90 -> 0.80 | Improved |
| add perception, prediction and decision CoTs to single-turn QA (ablation) | RDA-Driver · nuScenes val (UniAD metrics) | L2 avg (m): 1.01 -> 0.90 | Improved |
| drop any one of perception, prediction, decision CoT from multi-turn setup (ablation) | RDA-Driver · nuScenes val (UniAD metrics) | L2 avg (m): 0.80 -> 0.87 to 0.91 | Hurt |
| evaluate on DriveLM-nuScenes instead of nuScenes to reduce ego-status leakage | UniAD · UniAD metrics | L2 avg (m): 1.03 -> 3.00 | Hurt |
| RDA-Driver versus VAD on ST-P3 L2 | RDA-Driver · nuScenes val (ST-P3 metrics) | L2 avg (m): 0.37 -> 0.40 | No clear effect |
Closest papers by meaning
- Drive-R1: Bridging Reasoning and Planning in VLMs for Autonomous Driving with Reinforcement Learning (AAAI 2026 · similarity 0.91)
- DriveCombo: Benchmarking Compositional Traffic Rule Reasoning in Autonomous Driving (CVPR 2026 · similarity 0.90)
- nuReasoning: A Reasoning-Centric Dataset and Benchmark for Long-Tail Autonomous Driving (NeurIPS 2026 · similarity 0.89)
- Hybrid-Driving: An Autonomous Driving Decision Framework Integrating Large Language Models, Knowledge Graphs and Driving Rules (AAAI 2025 · similarity 0.89)
- OneVL: One-Step Latent Reasoning and Planning with Vision-Language Explanation (NeurIPS 2026 · similarity 0.89)
- More Than Meets the Eye? Uncovering the Reasoning-Planning Disconnect in Training Vision-Language Driving Models (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.
