Stochastic Characterization of Traction on Granular Terrain

Mobile robots traversing disaster sites, extraterrestrial landscapes, and other outdoor environments will encounter terrain covered with loose stones or rubble. When traversing such granular terrain, it will be important for the robot to have a model of available traction forces that can be used to plan trajectories and control the robot’s gait as it crosses the terrain. Interaction with granular terrain is inherently difficult to model due to the enormous number of dynamic interactions with and between the many small bodies involved. In this paper, we explore two probabilistic models of traction with data from experiments performed with a robotic foot interacting with a bed of gravel. These models include Tustin friction and Gaussian processes. We present the results of these two independent models, as well as a third method incorporating both.