Ego Motion Estimation of an Elastically Actuated Quadrupedal Robot: Fusion of Inertial Sensors and Forward Kinematics

Abstract

Stable locomotion in legged robots faces the challenges of intermittent ground contact of the feet and the possibility of slippage, especially when uneven terrain is traversed. The solution to these problems is the accurate, real-time estimation of the robot state, which additionally allows for navigation. For this purpose, different data may be collected from a variety of sensor modalities, each with its own shortcomings. The combination of these data within a sensor fusion framework provides a better state estimate than any of the sensors could deliver alone. Furthermore, the prospect of acquiring such data exclusively from high frequency sensors loaded on the robot itself implies autonomy and hence is particularly appealing. This thesis implements a state estimation algorithm for the quadruped robot Bert, with the sole use of on-board sensors. Specifically, IMU (acceleration and rotational rate) data is fused with kinematics measurements with the use of an Extended Kalman Filter. The algorithm is tested on Bert in a series of experiments featuring various conditions and the use of two different IMU sensors.