A comparison of search-based planners for a legged robot

Path planning for multi-DoF legged robots is a challenging task due to the high dimensionality and complexity of the planning space. We present our first attempt to build a path planning framework for the hydraulic quadruped - HyQ. Our approach adopts a similar strategy to [1], where planning is divided into a task-space and a joint-space part. The task-space planner finds a path for the center of gravity (COG) of the robot, while then the footstep planner generates the appropriate footholds under reachability and stability criteria. Next the joint-space planner translates the task-space COG trajectories into robot joint angles. We present a comparison of a set of search-based planning algorithms; Dijkstra, A* and ARA*, and evaluate these over a set of given terrains and a number of varying start and end points. All test runs support that our approach is a simple yet robust solution. We report comparisons in path length, computation time, and path cost, between the aforementioned planning algorithms.