Title: Optimizing robotic jumping on granular media

Author (Talk): Christian Hubicki, Georgia Institute of Technology

Abstract:

Substrate dynamics can have a significant impact on terrestrial locomotion behaviors; for example, movement strategies that perform well on hard ground can lead to failure on yielding substrates like granular media. We performed a robophysical experiment to investigate how granular media dynamics can be manipulated to modify performance in a simple behavior, jumping. Using recent models of granular bulk reaction forces during rapid intrusions [Aguilar & Goldman, Nature Physics, 2015] and a numerical optimal control algorithm, we control a jumping robot to leap from a bed of loose packed poppy seeds to a commanded apex using minimal motor work. The optimal control algorithm was able to exploit the terrain dynamics to jump to commanded heights; the patterns of robot self-deformation differed from those which produced similar heights on hard ground. For a subset of optimized behaviors, jumps overshot the commanded/predicted apex height by ~40%. In particular, this excessive jump height occurred when the intruding foot briefly (~40 ms) halted its penetration before resuming its downward thrust. As a result, we hypothesized the presence of additional un-modeled terrain dynamics which activate when the intruder pauses. Ongoing intrusion experiments with an instrumented robotic arm identify that short-duration pauses in motion (~100ms) result in increased resistive forces upon resumed intrusion. These experiments aim to identify further transient substrate physics which may be exploitable for the purpose of robotic locomotion.