A Minimum-Risk Path Planning Method Incorporating Bezier Curves in a Static Risk Field

This paper addresses the challenge of global path planning for vehicles within static risk fields in intelligent driving systems. To overcome the limited risk avoidance capabilities of traditional methods on structured roads, we propose a path generation framework that integrates risk gradient analysis with lane geometry constraints. Our approach comprises three key components: (1) local risk minimum point extraction based on eight-direction gradient analysis and non-maximum suppression; (2) distinct dual-path generation strategies for inside and outside intersections; and (3) a risk-constrained curve optimization method. Key innovations include a risk-field gradient-guided feature point extraction mechanism that enhances low-risk path point quality, two strategies tailored to complex intersection geometries for different risk scenarios, and a risk-aware curve parameter optimization method that balances safety and smoothness. Experiments conducted in complex intersections with traffic islands show that our method reduces the maximum and average risk values to 0.148 and 0.139, respectively, significantly outperforming real driving trajectories (with corresponding ranges of 0.193–0.276 and 0.149–0.179). Furthermore, the lower average DTW distance to real trajectories indicates higher behavioral similarity, while the path length (34.76 m) remains comparable to real trajectories (32.29–40.22 m). This provides a high-quality global reference path for intelligent vehicles that effectively balances safety with conformity to real driving habits.