Multi-risk Aware Trajectory Planning for Car-like Robot in Highly Dynamic Environments

Safe trajectory planning in highly dynamic environments remains a substantial challenge. Traditional risk-based trajectory planning algorithms solve planning problems in environments where humans and robots coexist, by taking into account the risk of collision with both dynamic and static obstacles. However, in highly dynamic environments, their safety during navigation is not guaranteed due to the singularity of risk assessment. Therefore, we first introduce a time-varying density risk map to represent external risks, then introduce a novel interact risk map generated based on the relative state of the robot and dynamic obstacles to represent interact risks. Finally, these two risks are combined based on a convex combination model to obtain a fusion risk map. The planner searches for an efficient, kinematically feasible, and safe trajectory by exploring the state graph built on motion primitives based on the fusion risk map. We have conducted a large number of simulation experiments in various challenging environments, and the experimental results have verified the effectiveness of the algorithm.