State Estimation for Legged Robots using Proprioceptive Sensors

Abstract

Mobile robots need good estimates of their state to perform closed-loop control in structured and unstructured environments. A number of existing algorithms rely on data fusion from multiple sensors to compute these estimates. This work focuses on state estimation using sensors which only measure information (acceleration, motor speed, joint angles) internal to the robot – proprioceptive sensors – since measurements of external features (light intensities, distance measurements, sound amplitude) may not always be reliable. Wheeled robots conventionally use IMUs and motor encoders for robust proprioceptive odometry. Legged robots, however, interact with their environment through intermittent foot-ground contacts which introduces additional noise in the IMU and joint encoder measurements making this problem challenging. We implement an Extended Kalman Filter (EKF) based state estimator which uses foot-ground contact information to counteract noisy sensor measurements from the IMU and motor encoders. This method has been implemented on simulation based quadruped and on an actual hexapod system.