HMSim: A Hierarchical Multi-Agent Simulator for Autonomous Vehicles
Haolan Liu, Jishen Zhao, Liangjun Zhang
AI summary
Problem
Most existing simulators prioritize sensor data plausibility while neglecting realistic driving behavior modeling, particularly struggling with behavioral diversity, rare near-collision scenarios, and error accumulation in long-term closed-loop rollouts.
Approach
The framework splits simulation into a high-level transformer-based controller that predicts diverse long-term driving intentions, and a low-level reinforcement learning policy that refines these goals into reactive, per-timestep actions using common-sense rewards.
Key results
- State-of-the-art prediction accuracy (minADE: 1.33m) on the Waymo dataset
- Enhanced behavioral diversity (MASD: 2.51m) with reduced collision and off-road rates
- Zero traffic light violations and superior scene plausibility in closed-loop runs
- Joint high-level regularization and low-level RL training outperforms prior baselines
Why it matters
Provides a scalable, safe testing environment for developing and evaluating robust autonomous driving policies by generating diverse, interactive, and realistic urban traffic scenarios.
Abstract
This paper addresses the challenge of developing a realistic urban-driving simulator to accurately model agent be- haviors, a crucial component for self-driving car development. Most previous simulators focus on the plausibility of sensor data synthesis, whereas the plausibility of driving behaviors is poorly explored. To tackle this problem, we propose a hierarchical ar- chitecture, which comprises (i) a high-level intention simulation summarizing driving scenarios and (ii) a low-level policy trained by reinforcement algorithms to refine plans. Unlike existing simulators, our approach captures diverse behaviors, even sub- optimal ones, vital for robust policy training and evaluation. We also highlight the importance of interactive simulations over static scenarios for realistic policy development. Exten- sive experiments demonstrate that our approach significantly improves long-term behavior prediction and closed-loop sim- ulation, enhancing the realism and diversity of urban-driving simulations. The videos of this work are available in our project page: https://sites.google.com/ucsd.edu/h-sim/home.