Intervention Feasible Region and Driver Risk Capacity Aware Human-Machine Collaborative Safe Trajectory Planning

The primary goal of human-machine collaborative driving is to improve driving safety, and the inconsistent tracking goals between humans and machines are one of the main factors causing safety accidents. Therefore, this article proposes a human-machine collaborative safe trajectory planning algorithm that considers the intervention feasible region (IFR) and the driver risk capacity (DRC), aiming to ensure driving safety from the planning perspective. On the one hand, the IFR solution and DRC assessment method are constructed. Based on the trajectory predicted by the constant turn rate and acceleration (CTRA) model and feasible boundary constraints, the IFR for the safe action space is solved. A method for quantifying the bidirectional subjective perceived risk is proposed to accurately assess personalized risk capacities. Provide an accurate decision-making basis for trajectory planning. On the other hand, a risk-adaptive safe deep reinforcement learning (DRL) trajectory planning algorithm, RAS- $\beta $ -SAC, is designed. Specifically, the beta distribution is introduced to adapt to the bounded safe action space defined by the IFR. At the same time, the risk relabeling module is designed to make the planned trajectory meet the personalized driving pReferences of drivers, thereby reducing human-machine conflicts and further improving trajectory safety. Experimental results show that the planned trajectory not only meets the safety constraints, but also satisfies the DRC, achieving dual safety guarantees of objective physical safety and driver's subjective acceptability.