Research map / Multi-agent systems and games / Human-AI and goal-conditioned
TrafficSim: Learning To Simulate Realistic Multi-Agent Behaviors
Check your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 heuristic car-following with jointly learned multi-agent behavior | TrafficSim · ATG4D | SCR₁₂s (%): 1.19 -> 0.50 | Helped |
| Replace heuristic car-following with jointly learned multi-agent behavior | TrafficSim · ATG4D | minSFDE (m): 4.97 -> 1.13 | Helped |
| Use TrafficSim rather than the lowest-TRV heuristic baseline | TrafficSim · ATG4D | TRV₁₂s (%): 0.25 -> 2.77 | Hurt |
| Train an imitation planner on TrafficSim-generated scenarios instead of the same amount of real data | Imitation planner · ATG4D planner evaluation | Planning L2 (m): 4.85 -> 4.52 | Helped |
| Add closed-loop training with backpropagation through simulation (ablation) | TrafficSim, one-step plan without common-sense loss (M₀ to M₂) · ATG4D | SCR₁₂s (%): 5.92 -> 1.28 | Helped |
| Add closed-loop training with backpropagation through simulation (ablation) | TrafficSim, 10-step plan without common-sense loss (M₁ to M₃) · ATG4D | SCR₁₂s (%): 2.32 -> 0.60 | Helped |
| Use a 10-step plan instead of a one-step plan during closed-loop training (ablation) | TrafficSim without common-sense loss (M₂ to M₃) · ATG4D | SCR₁₂s (%): 1.28 -> 0.60 | Helped |
| Add common-sense collision loss to closed-loop training with 10-step plans (ablation) | TrafficSim (M₃ to M*) · ATG4D | SCR₁₂s (%): 0.60 -> 0.50 | Helped |
| Add common-sense collision loss to closed-loop training with 10-step plans (ablation) | TrafficSim (M₃ to M*) · ATG4D | TRV₁₂s (%): 3.02 -> 2.77 | Helped |
| Use a 10-step plan instead of a one-step plan during closed-loop training (ablation) | TrafficSim without common-sense loss (M₂ to M₃) · ATG4D | minSFDE (m): 1.02 -> 1.21 | Hurt |
| Reduce inference frequency from 2Hz to 1Hz | TrafficSim · ATG4D multi-step update | Runtime (s): 0.83 -> 0.45 | Helped |
| Reduce inference frequency from 2Hz to 1Hz | TrafficSim · ATG4D multi-step update | SCR₁₂s (%): 0.50 -> 0.85 | Hurt |
Closest papers by meaning
- BehaviorGPT: Smart Agent Simulation for Autonomous Driving with Next-Patch Prediction (NeurIPS 2024 · similarity 0.91)
- Advancing Multi-agent Traffic Simulation via R1-Style Reinforcement Fine-Tuning (ICLR 2026 · similarity 0.91)
- LASIL: Learner-Aware Supervised Imitation Learning For Long-term Microscopic Traffic Simulation (CVPR 2024 · similarity 0.90)
- Trajeglish: Traffic Modeling as Next-Token Prediction (ICLR 2024 · similarity 0.90)
- Beyond Self-Play and Scale: A Behavior Benchmark for Generalization in Autonomous Driving (NeurIPS 2026 · similarity 0.90)
- SceneDiffuser++: City-Scale Traffic Simulation via a Generative World Model (CVPR 2025 · similarity 0.89)
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.