Research map / Multi-agent systems and games / Human-AI and goal-conditioned
Waymax: An Accelerated, Data-Driven Simulator for Large-Scale Autonomous Driving Research
Cole Gulino, Justin Fu, Wenjie Luo, George Tucker, Eli Bronstein, Yiren Lu, Jean Harb, Xinlei Pan, Yan Wang, Xiangyu Chen, John D Co-Reyes, Rishabh Agarwal and 10 more
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 route conditioning to behavior cloning (ablation) | BC (Bicycle Discrete) · WOMD planning | off-route rate (%): 2.31 -> 0.96 | Improved |
| add route conditioning to behavior cloning (ablation) | BC (Bicycle Discrete) · WOMD planning | off-road rate (%): 1.45 -> 1.11 | Improved |
| add route conditioning to behavior cloning (ablation) | BC (Bicycle Discrete) · WOMD planning | collision rate (%): 4.92 -> 4.59 | Improved |
| add route conditioning to behavior cloning (ablation) | BC (Bicycle Discrete) · WOMD planning | route progress ratio (%): 128.73 -> 129.84 | No clear effect |
| use discrete bicycle action space instead of continuous bicycle | BC · WOMD planning | collision rate (%): 11.20 -> 4.59 | Improved |
| use discrete delta action space instead of continuous delta | BC · WOMD planning | off-road/kinematic infeasibility: 4.14 -> 4.42 | Mixed |
| evaluate with IDM sim agents instead of log playback | DQN trained with Playback · WOMD planning | collision rate (%): 8.67 -> 4.91 | Improved |
| train with IDM sim agents instead of playback, evaluated on playback (ablation) | DQN · WOMD planning | collision rate (%): 8.67 -> 25.15 | Hurt |
| train with IDM instead of playback, evaluated against IDM (ablation) | DQN · WOMD planning | collision rate (%): 4.91 -> 6.50 | Hurt |
| run on V100 GPU with batch 16 instead of batch 1 repeated | Waymax Step · single-agent env | runtime (ms): 2.48 | Improved |
| GPU V100 instead of CPU, batch size 1 | Waymax Step · single-agent env | runtime (ms): 131 -> 0.75 | Improved |
| adapted Wayformer vs BC trained for the task | Wayformer · WOMD planning | collision rate (%): 4.59 -> 10.68 | Hurt |
Closest papers by meaning
- Advancing Multi-agent Traffic Simulation via R1-Style Reinforcement Fine-Tuning (ICLR 2026 · similarity 0.89)
- BehaviorGPT: Smart Agent Simulation for Autonomous Driving with Next-Patch Prediction (NeurIPS 2024 · similarity 0.89)
- Beyond Self-Play and Scale: A Behavior Benchmark for Generalization in Autonomous Driving (NeurIPS 2026 · similarity 0.89)
- DriveGPT: Scaling Autoregressive Behavior Models for Driving (ICML 2025 · similarity 0.87)
- AlignDrive: Aligned Lateral-Longitudinal Planning for End-to-End Autonomous Driving (NeurIPS 2026 · similarity 0.87)
- Scenario Dreamer: Vectorized Latent Diffusion for Generating Driving Simulation Environments (CVPR 2025 · 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.
