Research map / Label-efficient and robust learning / Recognition and long-tail
CorrectAD: A Self-Correcting Agentic System to Improve End-to-end Planning in Autonomous Driving
Enhui Ma, Lijun Zhou, Tao Tang, Jiahuan Zhang, Junpeng Jiang, Zhan Zhang, Dong Han, Kun Zhan, Xueyang Zhang, Xianpeng Lang, Haiyang Sun, Xia Zhou and 2 more
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 DriveSora generated data (without PM-Agent) to random duplication (ablation) | CorrectAD on UniAD · nuScenes val | avg collision rate (%): 0.35 -> 0.31 | Improved |
| add PM-Agent to tailor data to failure cases (without DriveSora) (ablation) | CorrectAD on UniAD · nuScenes val | avg collision rate (%): 0.35 -> 0.29 | Improved |
| combine PM-Agent and DriveSora (ablation) | CorrectAD on UniAD · nuScenes val | avg L2: 1.09 -> 0.98 | Improved |
| use multi-round inquiry (classify then analyze) instead of one-step VLM prompt (ablation) | PM-Agent · Failure cause analysis | semantic distance: 4.72 -> 3.49 | Improved |
| use DriveSora instead of Panacea as video generator (ablation) | CorrectAD on UniAD · nuScenes val | avg collision rate (%): 0.27 -> 0.19 | Improved |
| increase CorrectAD iterations from 1 to 3 (ablation) | CorrectAD on UniAD · nuScenes val | avg L2: 1.06 -> 0.98 | Improved |
| increase CorrectAD iterations from 1 to 3 (ablation) | CorrectAD on UniAD · nuScenes val | Hellinger distance to failure distribution: 0.15 -> 0.09 | Improved |
| add multimodal prompt (layout and caption conditions) to DriveSora (ablation) | DriveSora · nuScenes val | NDS: 25.23 -> 36.37 | Improved |
| add multi-view spatial attention to DriveSora (ablation) | DriveSora · nuScenes val | CLIP: 79.5 -> 86.73 | Improved |
| add unconditional background nullification to CFG (CFG Text,Fore,Back vs Text,Fore) (ablation) | DriveSora CFG · nuScenes val | FVD / Map mIoU: FVD 89.12 -> 94.60; mIoU 34.40 -> 35.96 | Mixed |
| use multi-condition CFG instead of MagicDrive CFG (ablation) | DriveSora · nuScenes val | FVD: 164.48 -> 94.60 | Mixed |
| generate data with CorrectAD instead of retrieving data with AIDE | UniAD · nuScenes val | avg collision rate (%): 0.28 -> 0.19 | 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.
- ST-P3 — ST-P3: End-to-End Vision-Based Autonomous Driving via Spatial-Temporal Feature Learning (ECCV 2022) · better on 2 of 2 shared results
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
- DriveLaW: Unifying Planning and Video Generation in a Latent Driving World (CVPR 2026 · similarity 0.89)
- Scenario Dreamer: Vectorized Latent Diffusion for Generating Driving Simulation Environments (CVPR 2025 · similarity 0.87)
- DriveDreamer4D: World Models Are Effective Data Machines for 4D Driving Scene Representation (CVPR 2025 · similarity 0.87)
- AlignDrive: Aligned Lateral-Longitudinal Planning for End-to-End Autonomous Driving (NeurIPS 2026 · similarity 0.86)
- Generating Multimodal Driving Scenes via Next-Scene Prediction (CVPR 2025 · similarity 0.86)
- DriveTransformer: Unified Transformer for Scalable End-to-End Autonomous Driving (ICLR 2025 · similarity 0.86)
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.
