TY - JOUR
T1 - Collaborative robot dynamics with physical human–robot interaction and parameter identification with PINN
AU - Yang, Xingyu
AU - Zhou, Zhengxue
AU - Li, Leihui
AU - Zhang, Xuping
PY - 2023/11
Y1 - 2023/11
N2 - Collaborative robots are increasingly being used in dynamic and semi-structured environments because of their ability to perform physical Human–Robot Interaction (pHRI) to ensure safety. Therefore, it is crucial to model the dynamics of collaborative robots during pHRI to gain valuable insights into the system’s behavior when in contact with humans. In this work, a generic dynamic model is proposed for the quasi-static contact phase of pHRI, which considers the interaction dynamics and the complete structural dynamics of the collaborative robot. Moreover, a hybrid physics-informed neural network (PINN) is proposed, which utilizes a recurrent neural network (RNN) and the Runge–Kutta method to identify the joint dynamic parameters without complicated regressor construction. Experiments are conducted using a UR3e collaborative robot, and the PINN is trained using the acquired data. The results demonstrate the effectiveness of the PINN in identifying joint dynamics without prior knowledge, and the dynamic simulation of pHRI is consistent with the experimental results. The proposed model and PINN-based identification approach have the potential to improve safety and productivity in industrial environments by facilitating the control of pHRI.
AB - Collaborative robots are increasingly being used in dynamic and semi-structured environments because of their ability to perform physical Human–Robot Interaction (pHRI) to ensure safety. Therefore, it is crucial to model the dynamics of collaborative robots during pHRI to gain valuable insights into the system’s behavior when in contact with humans. In this work, a generic dynamic model is proposed for the quasi-static contact phase of pHRI, which considers the interaction dynamics and the complete structural dynamics of the collaborative robot. Moreover, a hybrid physics-informed neural network (PINN) is proposed, which utilizes a recurrent neural network (RNN) and the Runge–Kutta method to identify the joint dynamic parameters without complicated regressor construction. Experiments are conducted using a UR3e collaborative robot, and the PINN is trained using the acquired data. The results demonstrate the effectiveness of the PINN in identifying joint dynamics without prior knowledge, and the dynamic simulation of pHRI is consistent with the experimental results. The proposed model and PINN-based identification approach have the potential to improve safety and productivity in industrial environments by facilitating the control of pHRI.
KW - Physical human–robot interaction
KW - Collaborative robot
KW - Dynamic modeling
KW - Joint flexibility
KW - Parameter identification
KW - Physics-informed neural network
UR - http://www.scopus.com/inward/record.url?scp=85165388053&partnerID=8YFLogxK
U2 - 10.1016/j.mechmachtheory.2023.105439
DO - 10.1016/j.mechmachtheory.2023.105439
M3 - Journal article
SN - 0094-114X
VL - 189
JO - Mechanism and Machine Theory
JF - Mechanism and Machine Theory
M1 - 105439
ER -