Self-Supervised Terrain Classification for Planetary Surface

Exploration Rovers 2 3 4 Christopher A. Brooks, Karl Iagnemma 5 Department of Mechanical Engineering 6 Massachusetts Institute of Technology 7 Cambridge, MA 02139 8 [email protected], [email protected] 9 10 Abstract 11 12 For future planetary exploration missions, improvements in autonomous rover 13 mobility have the potential to increase scientific data return by providing safe 14 access to geologically interesting sites that lie in rugged terrain, far from landing 15 areas. To improve rover-based terrain sensing, this paper proposes a self16 supervised learning framework that will enable a robotic system to learn to predict 17 mechanical properties of distant terrain, based on measurements of mechanical 18 properties of similar terrain that has been previously traversed. In this framework, 19 a proprioceptive terrain classifier is used to distinguish terrain classes based on 20 features derived from rover-terrain interaction, and labels from this classifier are 21 used to train an exteroceptive (i.e. vision-based) terrain classifier. Once trained, 22 the vision-based classifier is able to recognize similar terrain classes in stereo 23 imagery. This paper presents two distinct proprioceptive classifiers—one based 24 on wheel vibration and one based on estimated traction force—as well as a vision25 based terrain classification approach suitable for environments with unexpected 26 appearance. The high accuracy of the self-supervised learning framework and its 27 supporting algorithms is demonstrated using experimental data from a four28 wheeled robot in an outdoor, Mars-analog environment. 29 30