Neural kinematics net for a redundant robot arm

Abstract

Neural net mechanisms are proposed for the geometric representation of the mapping operation of desired 2D trajectories into actuator movements for a four-joint machine (4JM). The 4JM is capable of drawing various trajectories on a desk surface (40 cm*40 cm)-similar to horizontal drawing movements of a human arm-with the vertical projection of the stationary shoulder joint as center of the reference system. In this redundant robot system the four actuators for rotary movements above the joints (shoulder, virtual joint, elbow, and hand) simultaneously receive control signals from the neural kinematics net (NKN). The goal of NKN is to move the fingertip of the 4JM in real time along a desired 2D trajectory as specified by another neural net module, the neural space net.<>