A Graph Robot Network for Force Observer of Teleoperation Systems

Abstract

Force information is crucial to teleoperation systems, which allows experts to perform tasks safely and reduces potential risks. However, due to the limitations of robot size and working environment, it is difficult to obtain contact force information through traditional force sensors. Therefore, this article proposes a graph robot network force observer, which can achieve sensorless force estimation without solving the robot dynamics. First, different association graphs are constructed according to the relative spatial position of the robot joints, establishing information transmission channels between each joint. Second, a neural reinforcement factor is introduced from robot joint motion information to enhance the feature information of different association subgraphs. Finally, a spatial and temporal feature fusion module is developed to extract and fuse both temporal and spatial information, which outputs the force information on the robot end effector along axes X, Y, and Z. Experiments in different environments and ablation experiments are carried out to verify the stability and generalization ability of the model. The experimental results show that the proposed force observer can accurately achieve the contact force between the robot and the object, and the proposed force observer achieves state-of-the-art performance compared with other algorithms.