A new learning method utilizing a zone approach for performance enhancement of joint controlled manipulators.

Abstract

The conventional design strategy of robot manipulator fixed gain joint controllers is to critically damp the system for the most extreme configuration condition to prevent overshoot. This results in a highly overdamped system for all other configuration conditions which leads to slow response and poor high speed performance. A "zone learning method" has been developed which entails partitioning the robot work volume into zones based on the positions of the links. The link velocity feedback gain is then modified by multiplier values associated with each zone. The values of the multipliers are determined by a learning process based on the performance criteria of fastest possible response without overshoot. This method can also be implemented using link velocity zones. The method results in significant improvement in the performance of individual joint controlled manipulators and can easily and inexpensively be added to the hardware of existing robots using similar controllers.