Scalable Game-Theoretic Racing via Opponent-Specific SE-IBR

This paper presents a scalable real-time game-theoretic planner for autonomous racing. An opponent-specific Sensitivity-Enhanced Iterative Best Response (SE-IBR) scheme is proposed, where each agent couples strategically with only one most-relevant rival, selected by overtaking and threat-based criteria, while maintaining hard collision constraints with all others. By including the sensitivity for the tagged opponent only, the controller update achieves$O(1)$complexity in the number of agents. This reduction enables faster and more stable iterations compared to all-opponent sensitivity baselines. Multivehicle racing simulations confirm that the method preserves competitive behaviors such as overtaking and blocking, ensures convergence, and substantially reduces per-iteration solve time. These results highlight the practicality of opponent-specific SEIBR as a scalable framework for high-speed, interactive racing.