Environment Identification for a Running Robot Using Inertial and Actuator Cues

Abstract

In this paper, we explore the idea of using inertial and actuator information to accurately identify the environment of an amphibious robot. In particular, in our work with a legged robot we use internal sensors to measure the dynamics and interaction forces experienced by the robot. From these measurements we use simple machine learning methods to probabilistically infer properties of the environment, and therefore identify it. The robot's gait can then be automatically selected in response to environmental changes. Experimental results show that for several environments (sand, water, snow, ice, etc.), the identification process is over 90 per cent accurate. The requisite data can be collected during a half-leg rotation (about 250 ms), making it one of the fastest and most economical environment identifiers for a dynamic robot. For the littoral setting, a gaitchange experiment is done as a proof-of-concept of a robot automatically adapting its gait to suit the environment.