Bipedal Robotic Running on Stochastic Discrete Terrain

Abstract

To navigate over discrete terrain with large displacements in the stepping locations, bipedal robots have to be able to perform agile and dynamic maneuvers such as jumping or running while also satisfying strict constraints on foot placement and ground contact forces. In this paper, we analyze the problem of bipedal running over stochastically varying discrete terrain with large changes in step lengths. Specifically, our method is based on designing a library of running gaits that are two-step-periodic. We illustrate the capabilities of the proposed controller through numerical simulations of a five link underactuated robot RABBIT, running over discrete terrain with step lengths that vary between 0.6m and 1.2m. This is about 1.5 times the robot's leg lengths and twice the step length that could have been achieved by walking.