Research map / Reinforcement learning / Imitation learning
CoIRL-AD: Collaborative-Competitive Imitation-Reinforcement Learning in Latent World Models for Autonomous Driving
Xiaoji Zheng, Ziyuan Yang, Yanhao Chen, Yuhang Peng, Yuanrong Tang, Gengyuan Liu, Bokui Chen, Jiangtao Gong
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 RL actor with latent world model and competition on top of LAW | LAW · nuScenes | avg L2 x Col: 0.15 -> 0.11 | Improved |
| add competition between IL and RL actors (ablation) | CoIRL-AD · nuScenes | avg L2 x Col: 0.21 -> 0.11 | Improved |
| decouple IL and RL into separate actors rather than merging losses (ablation) | CoIRL-AD · nuScenes | avg L2 x Col: 0.17 -> 0.11 | Improved |
| use two-stage IL then RL fine-tuning (ablation) | LAW · nuScenes | avg L2 x Col: 0.15 -> 18.23 | Hurt |
| use inverse causal mask in planning head (ablation) | CoIRL-AD · nuScenes | avg L2 x Col: 0.18 (AR) -> 0.11 | Improved |
| use inverse causal mask on pure IL baseline (ablation) | LAW · nuScenes | avg L2 x Col: 0.15 -> 0.18 | Hurt |
| use mismatched masks for IL and RL actors (ablation) | CoIRL-AD · nuScenes | avg L2 x Col: 0.11 -> 0.15 | Hurt |
| add latent world model to RL imagination (ablation) | CoIRL-AD · nuScenes | avg L2 x Col: 0.13 -> 0.11 | Improved |
| add small behavior cloning regularizer (beta 0.005) (ablation) | RL actor · nuScenes | avg L2 x Col: 0.13 -> 0.11 | Improved |
| train RL only without IL (ablation) | RL actor · nuScenes | avg L2 x Col: 0.15 -> 32.29 | Hurt |
| use CQL pure offline RL instead of world-model RL (ablation) | CoIRL-AD · nuScenes | avg L2 x Col: 0.11 -> 15.87 | Hurt |
| use PPO-lite instead of group-sampled policy gradient (ablation) | CoIRL-AD · nuScenes | avg L2 x Col: 0.11 -> 0.10 | No clear effect |
Closest papers by meaning
- RLfOLD: Reinforcement Learning from Online Demonstrations in Urban Autonomous Driving (AAAI 2024 · similarity 0.89)
- RAD: Training an End-to-End Driving Policy via Large-Scale 3DGS-based Reinforcement Learning (NeurIPS 2025 · similarity 0.88)
- Distill to Think, Foresee to Act: Cognitive-Physical Reinforcement Learning for Autonomous Driving (NeurIPS 2026 · similarity 0.87)
- Raw2Drive: Reinforcement Learning with Aligned World Models for End-to-End Autonomous Driving (in CARLA v2) (NeurIPS 2025 · similarity 0.87)
- GoIRL: Graph-Oriented Inverse Reinforcement Learning for Multimodal Trajectory Prediction (ICML 2025 · similarity 0.87)
- Dual RL: Unification and New Methods for Reinforcement and Imitation Learning (ICLR 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.
