Automation department seminar

Safety-critical multi-agent Monte Carlo tree search for mixed traffic coordination at intersections without traffic lights

Signal-free intersections are one of the most challenging scenarios for autonomous vehicles, especially in mixed traffic where VAs must coordinate their actions with unpredictable human drivers.

This presentation outlines a novel decision framework that integrates k-level game theory and Monte Carlo Tree Search (MCTS) to enable safe and efficient coordination between multiple vehicles. Our approach addresses three major challenges: (1) the computational explosion of joint action spaces in multi-agent environments, (2) the need to model the heterogeneous behaviors of autonomous vehicles and human drivers, and (3) real-time safety guarantees.

This presentation presents a new decision-making framework that integrates k-level game theory and Monte Carlo tree search to enable safe and efficient multi-vehicle coordination.

Thanks to a dual-filtering dynamic interaction graph, we reduce computational complexity from O(M^N) to O(M^n₀ + M^n₁ + M^n₂), speeding up computation by 60% while maintaining a zero collision rate in pure autonomous vehicle scenarios.

The framework has been validated by extensive simulations showing a 40-62% improvement in trajectories and robust performance in mixed traffic with 50% human drivers. This work bridges the gap between theoretical game theory and practical applications of autonomous driving.

The framework has been validated by extensive simulations showing trajectory improvement and robust performance in mixed traffic with human drivers.

Zhihao Lin is a PhD student in Autonomous Systems and Connectivity at the University of Glasgow.

His research focuses on multi-agent decision-making, structure-sensitive reinforcement learning and perception-integrated autonomy.

He has published 10 papers in leading IEEE journals (h-index: 9, https://scholar.google.com/citations?user=zyQOwasAAAAJ&hl=en) on autonomous driving, SLAM and interactive decision-making.

He is currently a visiting PhD student at LAMIH UMR CNRS 8201, in France.

He is a reviewer for IEEE Transactions on Intelligent Transportation Systems and other leading journals.