Neurocontrol of cooperative dual robot manipulators

An approach to the control of cooperative dual manipulator systems using an artificial neural network based control strategy is presented. The neural network based method augments the decentralized local joint linear control of two SCARA type IBM 7540 robots and greatly improves systems performance for position-position control. In the position-position control strategy, the objective of the controller is to ensure that the end-effectors of both manipulators accurately track the pre-specified Cartesian space trajectories despite the effects of uncontrolled time-varying interaction forces exerted through a common load. A content addressable memory type strategy based on Albus' Cerebellar Model Articulation Controller (albus, 1975) is used for the implementation of the neural controller. The neural network based controller is used to iteratively learn the characteristics of the system rather than having to prespecify an explicitly system model. The system is linearly controlled at the beginning of the iterative process, but as the task error decreases through iterations, the neural network controller takes over from the linear controller. The use of the linear controller in parallel with the neural controller ensures reasonable system performance during the early stages of neural controller learning. The net joint RMS positioners for the two robots decreases from 0.2 radian at the start of the iterations (mostly linear control) to 0.001 radian at the end of the iterations (mostly neural network control). The controllers themselves are composed of simple structures that are computationally efficient and suitable for real-time implementation with high sampling rates. Simulation results are provided to illustrate the proposed control strategy.

Duke Scholars

Author Devendra P. Garg Thomas Lord Department of Mechanical Engineering and Materia ...