Proprioceptive sensing for autonomous self-righting on unknown sloped planar surfaces

Abstract

Robots that operate in dynamic, unknown environments occasionally require error recovery methods to return to a preferred orientation for mobility (i.e. self-righting), thus preventing mission failure and enabling asset recovery. In this paper, we reduce to practice our previously developed framework for determining self-righting solutions for generic robots on sloped planar surfaces. We begin by briefly reviewing our framework. We then describe the development of a modular robot for examining the effectiveness of our framework. This robot utilizes only joint encoders and an inertial measurement unit (IMU) for sensing. Next, we test the fidelity of our sensors by comparing commanded values, sensor data, and ground truth as given by a Vicon motion capture sensor environment, yielding a baseline margin of error. We utilize this data to explore the robot's ability to determine unknown ground angles using only proprioceptive sensors in combination with a conformation space map, which is pre-computed using our framework. We then investigate the robot's ability to develop its own conformation space map experimentally, and compare it to the pre-computed map. Finally, we demonstrate the robot's ability to self-right on various ground angles using 1, 2, and 3 degrees of freedom.