Research map / Reinforcement learning / Visual and representation RL
End-to-End Urban Driving by Imitating a Reinforcement Learning Coach
Zhejun Zhang, Alexander Liniger, Dengxin Dai, Fisher Yu, Luc Van Gool
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 |
|---|---|---|---|
| replace Autopilot with Roach as the expert (ablation) | CILRS+DAGGER with L1 action loss · NoCrash-busy new town & new weather | driving score (%): 43 -> 66 | Improved |
| add action-distribution KL loss instead of L1 action loss (ablation) | CILRS+DAGGER supervised by Roach · NoCrash-busy new town & new weather | driving score (%): 66 -> 79 | Improved |
| add latent feature matching loss with command-augmented input (ablation) | CILRS+DAGGER with KL loss · NoCrash-busy new town & new weather | driving score (%): 79 -> 88 | Improved |
| add latent feature matching loss with command-augmented input (ablation) | CILRS+DAGGER with KL loss · NoCrash-busy new town & new weather | success rate (%): 74 -> 87 | Improved |
| add feature matching without command input on NoCrash (ablation) | CILRS with KL loss · NoCrash | driving score (%) | Hurt |
| add feature matching on LeaderBoard (ablation) | CILRS with KL loss · offline LeaderBoard | driving score (%) | Improved |
| add KL action distribution loss instead of L1 action loss (ablation) | CILRS supervised by Roach · offline LeaderBoard | driving score (%) | No clear effect |
| add value supervision on top of feature matching (ablation) | CILRS with KL and feature loss · NoCrash-busy | driving score (%) | Improved |
| add value supervision without feature matching (ablation) | CILRS with KL loss · NoCrash-busy | driving score (%) | No clear effect |
| add Beta distribution to PPO (ablation) | Roach (PPO+exp) · NoCrash-dense new town & new weather | success rate (%): 77 -> 87 | Improved |
| add exploration loss to PPO with Beta (ablation) | Roach (PPO+beta) · NoCrash-dense new town & new weather | success rate (%): 87 -> 83 | No clear effect |
| add exploration loss to PPO with Beta (ablation) | PPO+beta · CARLA training stability | episode return | Improved |
Closest papers by meaning
- Prioritizing Perception-Guided Self-Supervision: A New Paradigm for Causal Modeling in End-to-End Autonomous Driving (NeurIPS 2025 · similarity 0.87)
- RAD: Training an End-to-End Driving Policy via Large-Scale 3DGS-based Reinforcement Learning (NeurIPS 2025 · similarity 0.87)
- Raw2Drive: Reinforcement Learning with Aligned World Models for End-to-End Autonomous Driving (in CARLA v2) (NeurIPS 2025 · similarity 0.87)
- Feedback-Guided Autonomous Driving (CVPR 2024 · similarity 0.86)
- MAPLE: Latent Multi-Agent Play for End-to-End Autonomous Driving (NeurIPS 2026 · similarity 0.85)
- Knowledge Distillation from Single-Task Teachers to Multi-Task Student for End-to-End Autonomous Driving (AAAI 2024 · similarity 0.85)
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.
