RNN for Motion-Force Control of Redundant Manipulators with Optimal Joint Torque [chapter]

Xuefeng Zhou, Zhihao Xu, Shuai Li, Hongmin Wu, Taobo Cheng, Xiaojing Lv
2020 AI based Robot Safe Learning and Control  
Precise position force control is the core and difficulty of robot technology, especially for robots with redundant degrees of freedom. For example, track-based controls often fail to grind the robot due to the intolerable impact force applied to the end-effector. The main difficulties lie in the coupling of motion and contact forces, redundancy analysis and physical constraints. In this chapter, we propose a new motion force control strategy under the framework of recursive neural network. The
more » ... tracking error and contact force are described respectively in the orthogonal space. By choosing the minimum joint torque as the secondary task, the control problem is transformed into the QP problem under multi-constraint conditions. In order to obtain real-time optimization of the joint torque relative to the non-convex joint angle, the original QP is reconstructed at the velocity level, and the original objective function is replaced by the time derivative. Then a convergent dynamic neural network is established to solve the improved QP problem online. The robot position control based on recursive neural network is extended to the robot position control based on position force, which opens a new way for the robot to turn from simple control angle to crossover design with convergence and optimality. Numerical results show that the proposed method can realize precise position force control, deal with inequality constraints such as joint angular velocity and torque limitation, and reduce joint torque consumption by 16% on average.
doi:10.1007/978-981-15-5503-9_6 fatcat:pouof2e5hbatdftsyauvbufjum