State estimation for legged robots

Abstract

This report investigates the effectiveness of the current state estimation algorithm for quadruped robots that is suitable in a non-GPS denied environment, focusing on the application of Kalman Filters (Linear and Extended Kalman Filter). The primary objective is to evaluate its performance and compare it with the newly implemented algorithms. In this report, Unitree’s Go1 quadruped robot is employed, utilizing its on-board sensors such as Inertia Measurement Unit, joint encoders and foot force sensors within the Kalman Filter framework. Additionally, supplementary cameras are incorporated by mounting them onto the robot to analyse their impact on the estimated state output. Furthermore, data collection was performed on two scenarios: flat ground and on the stairs, to test the robustness of the methods. To evaluate the effectiveness of each algorithm, the estimated state from various methods is being compared with the actual trajectory which was obtained from the OptiTrack cameras. Through comprehensive evaluation, it was found that the implemented algorithm that had visual aid (LKF-VILO) produced the most favourable outcome obtaining the lowest Root Mean Square Error in all 3 dimensions. These findings underscore the importance of state estimation and the need to further improve and advance in state estimation techniques for legged robotic systems, ensuring a much more efficient localization process.