Research map / Multi-agent systems and games / LLM agents
Improving Agent Behaviors with RL Fine-tuning for Autonomous Driving
Zhenghao Peng, Wenjie Luo, Yiren Lu, Tianyi Shen, Cole Gulino, Ari Seff, Justin Fu
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 closed-loop RL fine-tuning to pre-trained model | MotionLM 1M · WOSAC | Composite: 0.490 -> 0.597 | Improved |
| add closed-loop RL fine-tuning to pre-trained model | MotionLM 10M · WOSAC | Composite: 0.549 -> 0.608 | Improved |
| add RL fine-tuning | MotionLM 1M · WOSAC | Collision: 0.544 -> 0.863 | Improved |
| add RL fine-tuning | MotionLM 1M · WOSAC | ADE: 6.332 -> 2.436 | Improved |
| add behavioral-cloning pre-training before RL (ablation) | RL fine-tuned model · WOSAC | Composite: 0.320 -> 0.597 | Improved |
| add collision penalty (weight 2) to reward (ablation) | RL fine-tuning, 1M model · WOSAC | Collision: 0.834 -> 0.863 | Improved |
| add collision penalty (weight 2) to reward (ablation) | RL fine-tuning, 1M model · WOSAC | ADE: 2.405 -> 2.436 | Hurt |
| increase collision weight to 10 (ablation) | RL fine-tuning, 1M model · WOSAC | Collision: 0.863 -> 0.831 | Hurt |
| RL fine-tune sim agents for policy evaluation | MotionLM 10M · Policy evaluation vs log-playback | rank correlation: 0.845 -> 0.866 | Improved |
| RL fine-tune sim agents for policy evaluation | MotionLM 10M · Policy evaluation vs log-playback | absolute error: 10.014 -> 9.509 | Improved |
| fine-tune 10M model versus state-of-the-art MVTE | Fine-tuned 10M · WOSAC | Composite: 0.608 vs 0.645 | Hurt |
Closest papers by meaning
- Beyond Self-Play and Scale: A Behavior Benchmark for Generalization in Autonomous Driving (NeurIPS 2026 · similarity 0.91)
- Advancing Multi-agent Traffic Simulation via R1-Style Reinforcement Fine-Tuning (ICLR 2026 · similarity 0.91)
- BehaviorGPT: Smart Agent Simulation for Autonomous Driving with Next-Patch Prediction (NeurIPS 2024 · similarity 0.89)
- DriveGPT: Scaling Autoregressive Behavior Models for Driving (ICML 2025 · similarity 0.88)
- AdaWM: Adaptive World Model based Planning for Autonomous Driving (ICLR 2025 · similarity 0.88)
- RLFTSim: Realistic and Controllable Multi-Agent Traffic Simulation via Reinforcement Learning Fine-Tuning (CVPR 2026 · similarity 0.88)
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.
