Experimental Results for Dexterous Quadruped Locomotion Planning with RoboSimian

Abstract

RoboSimian is a quadruped robot inspired by an ape-like morphology, with four symmetric limbs that provide a large dexterous workspace and high torque output capabilities. Advantages of using RoboSimian for rough terrain locomotion include (1) its large, stable base of support, and (2) existence of redundant kinematic solutions, toward avoiding collisions with complex terrain obstacles. However, these same advantages provide significant challenges in experimental implementation of walking gaits. Specifically: (1) a wide support base results in high variability of required body pose and foothold heights, in particular when compared with planning for humanoid robots, (2) the long limbs on RoboSimian have a strong proclivity for self-collision and terrain collision, requiring particular care in trajectory planning, and (3) having rear limbs outside the field of view requires adequate perception with respect to a world map. In our results, we present a tractable means of planning statically stable and collision-free gaits, which combines practical heuristics for kinematics with traditional randomized (RRT) search algorithms. In planning experiments, our method outperforms other tested methodologies. Finally, real-world testing indicates that perception limitations provide the greatest challenge in real-world implementation.