Abstract:Closed-loop traffic simulation requires agents that are both scalable and behaviorally realistic. Recent self-play reinforcement learning approaches demonstrate strong scalability, but their equilibrium strategies fail to capture the socially aware behaviors of real human drivers. We propose a hierarchical architecture that goes beyond self-play by combining high-level multi-agent interaction reasoning with low-level continuous trajectory realization. Specifically, a Stackelberg-style Multi-Agent Reinforcement Learning (MARL) module generates interaction-aware intention commands. These commands condition a low-level continuous motion module, translating the strategic intent into physically consistent, scene-responsive control sequences. To mitigate distribution shift in closed-loop deployment, we introduce a hybrid co-training scheme combining MARL with auxiliary recovery supervision. Experiments on a SUMO-based urban network demonstrate that the proposed framework achieves superior control smoothness and safety compared to self-play and passive imitation baselines, while maintaining competitive traffic efficiency.
| Comments: | Submitted to IEEE Robotics and Automation Letters (RA-L) |
| Subjects: | Robotics (cs.RO); Artificial Intelligence (cs.AI) |
| Cite as: | arXiv:2605.09153 [cs.RO] |
| (or arXiv:2605.09153v1 [cs.RO] for this version) | |
| https://doi.org/10.48550/arXiv.2605.09153 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Weifan Zhang Mr [view email]
[v1]
Sat, 9 May 2026 20:31:34 UTC (1,034 KB)