A cooperative decision-making method for CAVs from the perspective of opinion dynamics

Autonomous vehicles face significant challenges when dealing with severe conflicts or complex weaving patterns. Though cooperative decision-making can address conflicts, concerns remain regarding the practicality and the substantial computational burden. Additionally, the neglect of social rules hinders effective interaction between connected automated vehicles (CAVs) and human-driven vehicles (HVs). To achieve efficient cooperation in mixed-traffic flows, this paper selects the unsignalized roundabout as the test site and proposes a negotiation decision-making framework based on opinion dynamics, which consists of opinion formation, quantification, consensus, and implementation. Specifically, the rights-of-way are considered as opinions and evaluated using utility. All CAVs achieve consensus on the allocation of the rights-of-way through communication, and ultimately accomplish motion control via rolling optimization. The opinion consensus process employs a heuristic algorithm based on entropy-iteration, effectively promoting convergence towards the optimal solutions and significantly reducing computational burden. In order to enhance the social compliance, a Support Vector Machine (SVM) model is constructed using naturalistic driving datasets. And the intentions of human drivers are predicted and integrated into the whole framework. Comparative tests against rule-based and game-based methods show that the proposed approach demonstrates superior performance in terms of safety, comfort, efficiency, and economy. Especially in the systemic cost, it can reduce average delays by 12.4 % to 29.8 %. Finally, hardware-in-the-loop (HIL) experiments are carried out, highlighting the potential for real-world application.