Neural redundant robotic trajectory optimization with diagnostic motor control

The joint commands of a simulated redundant PUMA robot and hand are allocated via a suggested command weight neural network for each DC motor. Three inputs of this network are diagnostic outputs of the neural network motor controller. Commands minimize time, energy expended, and error while attempting to reduce stress on motors experiencing extremes in loading, friction, or stiffness. This diagnostic control, combined with knowledge of a best vector of approach provided by an object grasp category network, enables the hand to attempt an accurate and stable first grasp of an object.<>