Dynamic optimization of redundant manipulators in worst case using recurrent neural networks

The aim of this paper is to find a comprehensive dynamic performance index (CDPI) for evaluating dynamic merit, and to develop a procedure for the optimization of dynamic performance for redundant manipulators in the worst case. CDPI stands for the maximum normalized joint driving torque and it can be minimized by linear programming. To obtain the minimum CDPI solution, a recurrent neural-network-based computational scheme is proposed for real time implementation. Robot configurations reached using the proposed planning algorithm can obtain the minimum joint driving torques. Numerical simulations have been carried out which illustrate good performance capability from the viewpoint of torque optimization.