Research map / Multi-agent systems and games / Human-AI and goal-conditioned
ECoSim: Data Efficient Fine-Tuning for Controllable Traffic Simulation
Yu-Hsiang Chen, WEI-JER Chang, Yi-Ting Chen, Masayoshi TOMIZUKA
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 sketch FiLM adapter trained on 1% paired data to VBD-CL | VBD-CL · WOSAC | mADE: 2.7223 -> 0.9645 | Improved |
| add sketch FiLM adapter trained on 1% paired data to SMART-tiny-CLSFT | SMART-tiny-CLSFT · WOSAC | mADE: 1.5044 -> 0.2561 | Improved |
| add sketch adapter to VBD-CL | VBD-CL · WOSAC | Meta score: 0.7186 -> 0.7432 | Improved |
| add text FiLM adapter to SMART-tiny-CLSFT | SMART-tiny-CLSFT · WOSAC | mADE: 1.5044 -> 1.3537 | Improved |
| FiLM adapter on ProSim with 1% data versus original ProSim with 100% data (sketch) | ProSim · Same-backbone ProSim comparison | mADE: 1.099 -> 1.130 | No clear effect |
| FiLM adapter on ProSim with 1% data versus original ProSim with 100% data (text) | ProSim · Same-backbone ProSim comparison | mADE: 2.330 -> 1.824 | Improved |
| use identity initialization versus random initialization for FiLM (ablation) | FiLM adapter · WOSAC sketch (SMART, 5,000 scenarios) | realism early in training | Improved |
| use multiplicative FiLM instead of additive modulation (ablation) | FiLM adapter · WOSAC sketch (SMART, 5,000 scenarios) | mADE / convergence | Improved |
| use dynamic (re-encode per timestep) conditioning instead of static for sketch (ablation) | FiLM adapter · WOSAC sketch control | mADE: 1.0598 -> 0.3970 | Improved |
| use dynamic instead of static conditioning for latent (ablation) | FiLM adapter · WOSAC latent control | mADE: 0.4842 -> 0.3151 | Improved |
| use context-matched retrieval instead of random agent selection for long-tail transfer (ablation) | Long-tail synthesis · Long-tail scenarios | collision rate: 0.3175 -> 0.1574 | Improved |
| restrict control signals to WOSAC target agents instead of all valid agents (sketch) (ablation) | FiLM adapter · WOSAC sketch control | mADE: 0.2561 -> 0.4602 | Hurt |
Closest papers by meaning
- RLFTSim: Realistic and Controllable Multi-Agent Traffic Simulation via Reinforcement Learning Fine-Tuning (CVPR 2026 · similarity 0.89)
- Causal Composition Diffusion Model for Closed-loop Traffic Generation (CVPR 2025 · similarity 0.88)
- Advancing Multi-agent Traffic Simulation via R1-Style Reinforcement Fine-Tuning (ICLR 2026 · similarity 0.87)
- Closed-Loop Supervised Fine-Tuning of Tokenized Traffic Models (CVPR 2025 · similarity 0.87)
- BehaviorGPT: Smart Agent Simulation for Autonomous Driving with Next-Patch Prediction (NeurIPS 2024 · similarity 0.87)
- SceneDiffuser++: City-Scale Traffic Simulation via a Generative World Model (CVPR 2025 · similarity 0.86)
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.
