Automatic Calibration Method of Leg Joint Angles of Quadruped Robot Based on Machine Vision

Abstract

Leg calibration of the quadruped robot is of great significance to improve the positioning accuracy. However, the encoder used to represent the real pose of the quadruped robot often exhibits zero-point drift, which reduces the positioning accuracy. The traditional calibration method is to manually calibrate each joint angle by technicians, which is time-consuming and labor-intensive. This paper proposes an online intelligent kinematics calibration method for quadruped robots based on machine vision and deep learning to simplify the calibration process and improve the calibration accuracy. The method includes two parts: identifying the marker fixed on the legs through target detection and calculating the center coordinates of the markers, and solving each joint angle by establishing an inverse kinematics neural network based on deep learning. We conducted a series of experiments to verify the accuracy of the method. The experimental results show that compared with the traditional manual calibration, the proposed method can greatly improve the calibration efficiency on the premise of meeting the calibration accuracy.