Research map / Reinforcement learning / Visual and representation RL
ZTRS: Zero-Human Demonstration End-to-end Autonomous Driving with Trajectory Scorer
Zhenxin Li, Nadine Chang, Wenhao Yao, Xinglong Sun, Zi Wang, Maying Shen, Jingde Chen, Jingyu Song, Kailin Li, Zuxuan Wu, Shiyi Lan, Jose M Alvarez
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 |
|---|---|---|---|
| train with rewards only, no human demonstrations (ZTRS vs GTRS-Dense) | trajectory scorer V2-99 · Navhard | EPDMS: 41.7 -> 45.5 | Improved |
| ZTRS vs GTRS-Dense | trajectory scorer EVA-ViT-L · Navhard | EPDMS: 43.4 -> 46.2 | Improved |
| ZTRS vs GTRS-Dense on closed-loop | trajectory scorer · HUGSIM | HD-Score: 28.2 -> 32.9 | Improved |
| ZTRS vs GTRS-Dense on nominal driving | trajectory scorer V2-99 · Navtest | EPDMS: 84.0 -> 85.3 | Improved |
| ZTRS vs DriveSuprim on nominal driving | trajectory scorer V2-99 · Navtest | EPDMS: 86.0 -> 85.3 | Hurt |
| exhaustive likelihood-form optimization (EPO) vs single best-trajectory imitation (ablation) | trajectory scorer V2-99 · Navtest | EPDMS: 76.7 -> 84.2 | Improved |
| add correction term b (temporal consistency) to the reward (ablation) | ZTRS V2-99 · Navtest | EC: 53.8 -> 77.2 | Improved |
| add correction term b to the reward (ablation) | ZTRS V2-99 · Navtest | EPDMS: 84.2 -> 85.3 | Improved |
| use EPO instead of REINFORCE (ablation) | trajectory scorer V2-99 · Navtest | EPDMS: 75.0 -> 85.3 | Improved |
| use EPO instead of REINFORCE across seeds (ablation) | trajectory scorer V2-99 · Navtest | EPDMS variance: 72.6/75.0 -> 85.3/85.5 | Improved |
| add progress reward to safety-only reward (ablation) | ZTRS V2-99 · Navtest | EPDMS: 72.0 -> 82.4 | Improved |
| add comfort rewards on top of safety+progress (ablation) | ZTRS V2-99 · Navtest | EPDMS: 82.4 -> 85.3 | Improved |
Later papers that compared with this method
1 later accepted paper reports ZTRS as a baseline in their results tables; 1 of them report a better number on most of the results they share with it.
- HAD: Combining Hierarchical Diffusion with Metric-Decoupled RL for End-to-End Driving (ECCV 2026 · 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
- RAD: Training an End-to-End Driving Policy via Large-Scale 3DGS-based Reinforcement Learning (NeurIPS 2025 · similarity 0.88)
- RLfOLD: Reinforcement Learning from Online Demonstrations in Urban Autonomous Driving (AAAI 2024 · similarity 0.87)
- DriveDPO: Policy Learning via Safety DPO For End-to-End Autonomous Driving (NeurIPS 2025 · similarity 0.87)
- Prioritizing Perception-Guided Self-Supervision: A New Paradigm for Causal Modeling in End-to-End Autonomous Driving (NeurIPS 2025 · similarity 0.86)
- Distill to Think, Foresee to Act: Cognitive-Physical Reinforcement Learning for Autonomous Driving (NeurIPS 2026 · similarity 0.86)
- Diffusion-Based Planning for Autonomous Driving with Flexible Guidance (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.
