Neuro-adaptive hybrid position/force control of robotic manipulators

This paper presents a neural network approach to the hybrid control of manipulators while interacting with the environment. The overall control strategy comprises a nominal model of the manipulator with separate neural network compensators along the force and motion controlled directions in the task co-ordinate frame. With the learning mechanism operating in the task space, modelling errors, dynamic friction and changes in environment stiffness are automatically compensated for, which result in highly desirable task oriented performance characteristics. Simulation results are provided using the PUMA 560 arm which demonstrates the applicability of the proposed method to the position/force hybrid control of manipulators