Online identification of inertial parameters of a robot with partially combined links using IMU sensing

• Main Authors: Chen, Chensheng, Zhang, Wenwen, Liu, Tao, Zhang, Zhengyuan, Lu, Wenhao, Wang, Lei, Zheng, Yuanjin, Lin, Zhiping
• Other Authors: School of Electrical and Electronic Engineering
• Format: Article
• Language: English
• Published: 2023
• Subjects: Engineering::Electrical and electronic engineering; Dynamics Modeling; Initial Position Optimization
• Online Access: https://hdl.handle.net/10356/171314
• Institution: Nanyang Technological University

Accurate calculation of joint acceleration online is critical for detecting robot collisions when used in inverse dynamics to calculate joint torques. The conventional method for calculating joint acceleration is to employ the twice differentiation based on encoder data, which suffers from the problem of causing joint acceleration with excessive noise. To address this problem, an extended Kalman filter (EKF) sensor fusion method is proposed in this study, which combines data from encoder and inertial measurement unit (IMU) sensors to estimate joint motion information accurately. In an inertial parameter identification experiment, the first three links of a seven-degree-of-freedom (DoF) robot remain stationary and unexcited, so that the results of the identification of the last four links will be affected by their initial positions. To examine the effect of the initial positions of the first three links without introducing an excessive number of variables, joints 4+5 and 6+7 were combined. Furthermore, to improve the accuracy of the calculated joint torques, the fmincon() function is used to optimize a constrained nonlinear multivariable equation containing the joint position of the first three links, and the inertial parameters of the combined links are determined using the recursive least squares algorithm. The simulation and experimental results demonstrate that the joint motion information estimated by EKF is more accurate than conventional differentiation based on encoder output. In addition, the inertial parameters of the two combined links are calculated using an online least squares algorithm, which is computationally more efficient and practical for real-world scenarios than the conventional offline least squares algorithm.