Active perception for legged locomotion: every step is an experiment

Abstract

A legged robot can use its legs not only for locomotion but also to identify the material properties and shape of the terrain it traverses. The author formulates active perception techniques based on force/torque sensing to identify terrain stiffness and surface friction, and he presents the results of applying the techniques to characterize terrain transversability. He also describes purposeful contact sensing to update geometric terrain maps in a process that might be called 'shape from kicking,' that is, he describes how a walking robot can actively conduct perceptual experiments by stepping on and kicking things. He also specifies how the results of the experiment may benefit a robot walking over rugged terrain.