📝 SummarXiv

Abstract

Decision-making at unsignalized roundabouts poses substantial challenges for autonomous vehicles (AVs), particularly in mixed traffic environments where AVs must coordinate safely with human-driven vehicles (HDVs). This paper presents a safety-critical multi-agent Monte Carlo Tree Search (MCTS) framework that integrates both deterministic and probabilistic prediction models to facilitate cooperative decision-making in complex roundabout scenarios. The proposed framework introduces three key innovations: (1) a hierarchical safety assessment module that systematically addresses AV-to-AV (A2A), AV-to-HDV (A2H), and AV-to-Road (A2R) interactions through dynamic safety thresholds and spatiotemporal risk evaluation; (2) an adaptive HDV behavior prediction scheme that combines the Intelligent Driver Model (IDM) with probabilistic uncertainty modeling; and (3) a multi-objective reward optimization strategy that jointly considers safety, efficiency, and cooperative intent. Extensive simulation results validate the effectiveness of the proposed approach under both fully autonomous (100% AVs) and mixed traffic (50% AVs + 50% HDVs) conditions. Compared to benchmark methods, our framework consistently reduces trajectory deviations across all AVs and significantly lowers the rate of Post-Encroachment Time (PET) violations, achieving only 1.0\% in the fully autonomous scenario and 3.2% in the mixed traffic setting.