One of the important applications for the resolution of redundant manipulators is torque optimization, due to the fact that the actuators used for driving a manipulator have finite power ratings. Nevertheless, not many algorithms have been proposed to accomplish this objective. A survey of the existing local torque optimization control schemes is given in this article. It will be shown that all of them either encounter instability problems for long trajectories or fail in certain cases. For remedying these problems, the authors present the Minimum Velocity Norm (MVN) method, which is the most common approach for kinematic redundancy resolution and has never been adopted for torque optimization by other researchers. Simulation results show that the simple MVN method is moderate for short movements and is stable for long movements. Also, the MVN method can be applied to cases that may not be accomplished by some other approaches. Therefore, the MVN method is better than the other existing approaches for torque optimization. © 1994 John Wiley & Sons, Inc.