Research map / Label-efficient and robust learning / Recognition and long-tail
Unraveling the Effects of Synthetic Data on End-to-End Autonomous Driving
Junhao Ge, Zuhong Liu, Longteng Fan, Yifan Jiang, Jiaqi Su, Yiming Li, Zhejun Zhang, Siheng Chen
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 2k synthetic samples to small real training set (8k) | VAD · Waymo val open-loop | planning L2 avg / CR avg: 1.39 -> 1.18; 0.52 -> 0.45 | Improved |
| add 2k synthetic samples to larger real training set (28k) | VAD · Waymo val open-loop | planning L2 avg / CR avg: 1.09 -> 1.04; 0.32 -> 0.25 | Improved |
| add adaptive kinematic model instead of bicycle model (ablation) | Scene Controller · Waymo reconstructed scenes, VAD proxy | planning L2 @3s (m): 2.43 -> 2.06 | Improved |
| add adaptive kinematic model instead of bicycle model (ablation) | Scene Controller · Waymo reconstructed scenes, VAD proxy | MapSeg AP 0.5: 0.170 -> 0.185 | Improved |
| add 3DRealCar assets with directional shadows instead of virtual vehicle assets | Scene Renderer · Waymo reconstructed scenes | Car AP: 38.4 -> 39.6 | Improved |
| replace real foreground and background by rendered scene (3DRealCar Rep.) | VAD perception · Waymo reconstructed scenes | Car AP: 44.1 -> 39.6 | Hurt |
| fine-tune with 6k synthetic samples | VAD · Closed-loop unseen scenes | route completion / vehicle collision: 42.01 -> 53.03; 7.55 -> 3.43 | Improved |
| fine-tune with 6k synthetic samples | GenAD · Closed-loop unseen scenes | route completion / vehicle collision: 44.87 -> 49.65; 7.62 -> 5.45 | Improved |
| fine-tune with synthetic data on turn scenes | VAD · Closed-loop unseen scenes (turn) | route completion: 25.71 -> 34.72 | Improved |
| fine-tune with synthetic data on reconstructed training scenes | VAD · Closed-loop reconstructed scenes | route completion: 29.88 -> 39.02 | Improved |
| add bidirectional interaction to synthetic traffic | Scene Controller · Waymo scenes | interaction rate | Improved |
Closest papers by meaning
- SceneCrafter: Controllable Multi-View Driving Scene Editing (CVPR 2025 · similarity 0.89)
- DriveDreamer4D: World Models Are Effective Data Machines for 4D Driving Scene Representation (CVPR 2025 · similarity 0.89)
- DrivingSphere: Building a High-fidelity 4D World for Closed-loop Simulation (CVPR 2025 · similarity 0.89)
- Bench2Drive: Towards Multi-Ability Benchmarking of Closed-Loop End-To-End Autonomous Driving (NeurIPS 2024 · similarity 0.88)
- SceneDiffuser: Efficient and Controllable Driving Simulation Initialization and Rollout (NeurIPS 2024 · similarity 0.87)
- SceneDiffuser++: City-Scale Traffic Simulation via a Generative World Model (CVPR 2025 · similarity 0.87)
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.
