Applying Kalman filtering on a Quadruped Robot

Abstract

Cheetah-Cub is an open-loop controlled quadruped robot. It is equipped with a 9DOF Inertial Measurement Unit (IMU) which is not used for its locomotion.One strategy to exploit the full potential of the robot would be to use sensor data in a closed-loop control to adapt the gait depending on the terrain or the wanted speed. However, in order to have relevant data, sensor output has to be properly filtered and processed.

This semester project aims at using Kalman filtering techniques to estimate instantaneous roll and pitch angles, as well as instantaneous forward acceleration of the robot while trotting. A first quaternion-based Kalman filter is implemented for the orientation estimation. A second one estimates forward acceleration using a simple locomotion model and the gravity-corrected accelerometer data.

Results show that the Kalman filter is indeed suitable for extracting relevant locomotion data and outperforms standard filters in doing it. Yet, given the demanding requirements of the task, hardware limitations significantly restrict its performance.