Real-time remodeling of granular terrain for robot locomotion
Abstract
Recent studies of robot movement in flowable granular media inspired by difficulties faced by extraterrestrial rovers reveal a coupled locomotor/substrate effect where the robot spontaneously remodels its environment. Such coupling occurs in certain limb/wheel movement patterns that results in a localized granular flow allowing the robot to effectively “swim” up highly flowable slopes. However, these gaits were discovered via trial and error by human operators. as the highly hysteretic nature of easily flowable terrain also creates tractability and predictability challenges in locomotion planning and gait policies. To overcome this, additional anchoring structures on intruding appendages can dynamically stabilize slopes to prevent undesired flows and slipping during locomotion. Granular media’s multiphase properties make it amenable to creative manipulations dependent on the physics of the intruding structure. A pair of robot studies showcase both selective solidification and fluidization strategies in flowable slopes to locomote successfully. To accelerate gait discovery in both studies, a machine learning approach for real-time characterization of the terrain flow could allow robots to control the flowable substrate for effective locomotion. A future neural network trained with sufficient spatiotemporal terrain data could predict granular flow with high accuracy and generality, augmenting gait learning with knowledge of the environment’s evolution during movement.
Research Areas
