Research map / Multi-agent systems and games / Human-AI and goal-conditioned
SimScale: Learning to Drive via Real-World Simulation at Scale
Haochen Tian, Tianyu Li, Haochen Liu, Jiazhi Yang, Yihang Qiu, Guang Li, Junli Wang, Yinfeng Gao, Zhang Zhang, Liang Wang, Hangjun Ye, Long Chen and 1 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 simulation data with pseudo-expert via sim-real co-training | LTF · navhard | EPDMS: 24.4 -> 30.2 | Improved |
| add simulation data with pseudo-expert via sim-real co-training | DiffusionDrive · navhard | EPDMS: 27.5 -> 32.6 | Improved |
| add simulation data with reward scoring | GTRS-Dense V2-99 · navhard | EPDMS: 41.9 -> 48.0 | Improved |
| add simulation data via co-training | LTF · navtest | EPDMS: 81.5 -> 84.4 | Improved |
| add simulation data via co-training | GTRS-Dense V2-99 · navtest | EPDMS: 84.0 -> 84.6 | Improved |
| use planner-based instead of recovery-based pseudo-expert (ablation) | planners · navhard scaling curves | EPDMS | Improved |
| scale simulation data to 1:1 sim-to-real with planner-based expert (ablation) | LTF (single-mode regression) · navhard | EPDMS | Hurt |
| scale simulation data with planner-based expert (ablation) | DiffusionDrive · navhard | EPDMS | Improved |
| use reward signal only in simulation instead of expert trajectories (ablation) | GTRS-Dense · navhard | EPDMS: V2-99 47.7 -> 48.0; ResNet34 46.1 -> 45.9 | Mixed |
| reward-only training on real data without real expert (ablation) | GTRS-Dense · navhard | EPDMS: 38.3 -> 37.6 | Hurt |
| use reactive instead of non-reactive simulation (ablation) | GTRS-Dense reward scoring · navhard | EPDMS: ResNet34 43.7 -> 44.4; V2-99 45.6 -> 46.7 | Improved |
| add a third reactive sampling round (ablation) | GTRS-Dense reward scoring · navhard | EPDMS: ResNet34 44.4 -> 45.0; V2-99 46.7 -> 47.9 | Improved |
Closest papers by meaning
- CausalDrive: Real-time Causal World Models for Autonomous Driving (ECCV 2026 · similarity 0.90)
- ReSim: Reliable World Simulation for Autonomous Driving (NeurIPS 2025 · similarity 0.89)
- Solving Motion Planning Tasks with a Scalable Generative Model (ECCV 2024 · similarity 0.89)
- Prioritizing Perception-Guided Self-Supervision: A New Paradigm for Causal Modeling in End-to-End Autonomous Driving (NeurIPS 2025 · similarity 0.88)
- SimGen: Simulator-conditioned Driving Scene Generation (NeurIPS 2024 · similarity 0.88)
- BehaviorGPT: Smart Agent Simulation for Autonomous Driving with Next-Patch Prediction (NeurIPS 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.
