A dual neural network is presented for real-time joint torque minimization of kinematically redundant manipulators, which corresponds to global kinetic energy minimization of robot mechanisms. Compared to other neural-network-based computation schemes for inverse kinematics, the proposed neural network with concise architecture is composed of only one single layer of neurons with the size equal to the dimensionality of the task workspace and is proven to be globally exponentially stable, which guarantees the sufficiently small end-effector tracking error. The proposed dual neural network has been simulated on six degrees of freedom (DOF) robot arm PUMA560 with the effectiveness and efficiency demonstrated.
展开▼
