Robust position/force control of multiple robots using neural networks

The control of multiple redundant robots, whose end-effectors grasp an object, involves complex control tasks. First, the multiple robotic system, for a cooperative task, forms closed kinematic chains that impose additional kinematic and dynamic constraints. Second, the interactive actions among the robots through the object lead to the essential need to control position and interactive force, simultaneously. Finally, the structured and unstructured uncertainties of the system may cause the system to be unstable. In this paper, a robust controller, which compensates the uncertainties of the dynamic system of the multiple robotic system, is presented in order to obtain good tracking performance of position and force, simultaneously, while satisfying the constraint conditions among the robots. A neural network architecture is proposed as one approach to the design and implementation of the robust controller. In particular, an on-line learning rule is provided for reportedly assigned tasks so that the system is robust to the structured/unstructured uncertainties; and the controller adjusts itself repeatedly to improve the performance progressively for each repeated task.