Legged Robot State Estimation using Invariant Kalman Filtering and\n Learned Contact Events

Abstract

This work develops a learning-based contact estimator for legged robots that\nbypasses the need for physical sensors and takes multi-modal proprioceptive\nsensory data as input. Unlike vision-based state estimators, proprioceptive\nstate estimators are agnostic to perceptually degraded situations such as dark\nor foggy scenes. While some robots are equipped with dedicated physical sensors\nto detect necessary contact data for state estimation, some robots do not have\ndedicated contact sensors, and the addition of such sensors is non-trivial\nwithout redesigning the hardware. The trained network can estimate contact\nevents on different terrains. The experiments show that a contact-aided\ninvariant extended Kalman filter can generate accurate odometry trajectories\ncompared to a state-of-the-art visual SLAM system, enabling robust\nproprioceptive odometry.\n