A mode map model to predict state transitions of multi-legged robots within obstacle fields

Environments filled with large obstacles such as rocks and boulders are challenging for legged robots. Inspired by the agile movements achievable by humans and animals through actively engaging obstacles, we (i) investigate how robot gaits affect robot dynamics during repeated contact with obstacles, and (ii) seek locomotion strategies that exploit the "disturbances" caused by robot-obstacle contacts to produce desired trajectories.

We studied a robophysics-informed model of a quadrupedal robot and found that---for periodic gaits and spatially periodic obstacle fields---robot behaviors often quickly become periodic, allowing the robot to traverse an obstacle field in different directions by switching between the periodic steady-states of different gaits, but without using any body-level steering. This reduced-order, discrete-time "mode map" model predicts the obstacle-modulated robot state transitions for a variety of gaits, aids in the systematic analysis of convergence mechanisms, and enables approximations of the basins of attraction of steady-state behaviors. Using these basins, we demonstrate an "obstacle-aided" navigation method that allows the robot to elicit obstacle collisions to generate desired trajectories.