Safety-Guaranteed Decision-Making and Planning for AVs Under Behavioral Uncertainties

With the widespread of autonomous vehicles(AVs), safety has become a major concern and raised much attention. Ensuring driving safety remains a challenge when autonomous vehicles(A V s) encounter scenarios with behavioral uncertainties, such as mandatory lane changing or emergency braking. To address this problem, this paper proposes a safety-guaranteed decision-making algorithm based on risk field and Monte-Carlo tree search(MCTS) with considering uncertain behaviors of surrounding vehicles(SVs). The algorithm consists of two components: risk assessment of SV s and decision-making for the ego vehicle(EV). Firstly, longitudinal and lateral safety distance functions are constructed, and an integrated risk assessment function is obtained. Secondly, the MCTS is adopted for the EV for decision-making and the corresponding trajectory is planned based decision waypoints. Finally, the proposed algorithm is validated in the real-world test dataset nuPlan. Results indicate that the proposed algorithm can effectively ensure driving safety.