Direct learning of control efforts for trajectories with different magnitude scales (Q1388113)

The paper introduces the concept of direct learning, defined as the direct generation of desired control profiles from different past control profiles, without repeated learning process. Direct learning control schemes are developed for some classes of nonlinear systems. The main advantages of these schemes result from the full utilization of the previously obtained control input signals with various magnitude scales and from the avoidance of repetitive learning. Simulation results for a single link robot manipulator illustrate the effectiveness of the proposed direct learning schemes.