Research map / Reinforcement learning / Visual and representation RL
Learning To Drive From a World on Rails
Dian Chen, Vladlen Koltun, Philipp Krähenbühl
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 camera augmentation (ablation) | Rails · NoCrash Dense (test town, test weather) | success rate (%): 42 -> 48 | Improved |
| add speed augmentation (ablation) | Rails · NoCrash Dense (test town, test weather) | success rate (%): 42 -> 58 | Improved |
| add both camera and speed augmentation (ablation) | Rails · NoCrash Dense (test town, test weather) | success rate (%): 42 -> 66 | Improved |
| add camera augmentation on top of speed augmentation (ablation) | Rails · NoCrash Dense (test town, train weather) | success rate (%): 64 -> 74 | Improved |
| add camera augmentation on top of speed augmentation (ablation) | Rails · NoCrash Dense (train town, test weather) | success rate (%): 80 -> 84 | Hurt |
| add auxiliary semantic segmentation loss (ablation) | Rails · NoCrash Empty (test town, test weather) | success rate (%): 46 -> 76 | Improved |
| add auxiliary semantic segmentation loss (ablation) | Rails · NoCrash Empty (train town, test weather) | success rate (%): 70 -> 100 | Improved |
| add factorized world model (F-DM instead of DM) (ablation) | Dreamer · CoRL17 / NoCrash | success rate (%): 37 -> 44 (Straight, train town); 0 -> 0 (Turn) | Mixed |
| add tabular Bellman distillation instead of CEM planning (ablation) | Rails · NoCrash Dense (test town, train weather) | success rate (%): 47 -> 74 | Improved |
| add world-on-rails factorization instead of full world model (Dreamer) (ablation) | Rails · CoRL17 Turn (train town) | success rate (%): 0 -> 100 | Improved |
| add Ornstein-Uhlenbeck noise to training trajectories (with both augmentations) (ablation) | Rails · NoCrash Dense (test town, test weather) | success rate (%): 66 -> 46 | No clear effect |
| add Rails instead of IA in traffic light compliance | Rails · NoCrash Dense (train town, train weather) | traffic light violations per hour: 15.41 -> 2.61 | Improved |
Closest papers by meaning
- ExploreVLA: Dense World Modeling and Exploration for End-to-End Autonomous Driving (ECCV 2026 · similarity 0.90)
- Driving in the Occupancy World: Vision-Centric 4D Occupancy Forecasting and Planning via World Models for Autonomous Driving (AAAI 2025 · similarity 0.89)
- AdaWM: Adaptive World Model based Planning for Autonomous Driving (ICLR 2025 · similarity 0.88)
- Prioritizing Perception-Guided Self-Supervision: A New Paradigm for Causal Modeling in End-to-End Autonomous Driving (NeurIPS 2025 · similarity 0.88)
- Navigation World Models (CVPR 2025 · similarity 0.87)
- Vista: A Generalizable Driving World Model with High Fidelity and Versatile Controllability (NeurIPS 2024 · 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.
