华中科技大学学报（自然科学版）2025，Vol.53Issue(11)：36-41,6.DOI:10.13245/j.hust.251141
基于ANN的机器人运动学误差标定方法
ANN-based calibration method for robot kinematic errors
摘要
Abstract
To improve the convergence speed and calibration accuracy in kinematic calibration of robots utilizing feed-forward artificial neural networks(ANNs),an ANN-based algorithm for calibrating robot kinematic error models was proposed.First,the angle information of the revolute joint was encoded using sine-cosine function and pairwise multiplication to integrate periodic information into the network.Then,kinematic error models were formulated based on both the body coordinate system and the base coordinate system.The network model's output was the pose matrix,enabling the fusion of the Jacobian matrix of the error model with that of the theoretical model through simple multiplication.This fusion facilitated the inverse solution without the necessity of separately learning forward and inverse kinematics.Experimental results show that encoding with sine-cosine and multiplication results in a twofold increase in the model's convergence speed,and a 66.7%reduction in error loss.In comparison to a fully black model,the position calibration error of the parallel error model decreases from 0.516 mm to 0.325 mm,marking a 37.02%reduction.关键词
机器人/运动学误差标定/微分运动学/前馈人工神经网络/正余弦旋转编码Key words
robotics/kinematic error calibration/differential kinematics/feed-forward artificial neural networks/sine-cosine rotation encoding分类
信息技术与安全科学引用本文复制引用
GAO Chao,PAN Bo,NIU Guojun,FU Yili..基于ANN的机器人运动学误差标定方法[J].华中科技大学学报（自然科学版）,2025,53(11):36-41,6.基金项目
浙江省尖兵研发攻关计划资助项目(2023C03010). （2023C03010）
