Kinodynamic Region Rapidly-exploring Random Trees (KRRRTs)

Kinodynamic motion planning is the problem of finding a collision-free path for a robot under constraints e.g., velocities and accelerations. State of the art techniques rely on sampling-based planning which samples and connects configurations until a valid path is found. Many sampling-based planners have been developed for non-holonomic problems e.g., Kinodynamic Rapidly-exploring Random Trees (KRRT). Although KRRT’s computational efficiency is competent, it lacks human problem solving skills. Human-guided planning combines automated planning and human intuition, allowing greater problem complexity and computational efficiency. A recent human-guided planner, Region Steering, allows sampling to bias or avoid user specified workspace regions, but was designed for holonomic bodies only. In this work, we extend Region Steering to handle such problems with dynamic constraints. Our novel method, Kinodynamic Region RRT (KRRRT), allows the user to specify attract or repel regions that the planner will concentrate or avoid planning, then applies KRRT’s motion constraints to the problem. In our experimental results, we found that utilizing regions made our planner up to six times faster than KRRT.