A method for planning minimum time joint trajectories for robot manipulators is discussed. The minimum time trajectory planning problem for manipulators is one of the minimum time control problems of non-linear systems. The optimal input torque/force is of a bang-bang type, except for the singular control derived from the Maximum Principle. An algorithm for generating a bang-bang control is proposed. In the algorithm, the switching time vector is updated to decrease the final state error. The proposed algorithm is applied to a simple manipulator with two links, and the solution by this algorithm is compared with the sub-optimal solution obtained by another approximate method.