The project goal is the development of a controller design for systems with unknown nonlinear dynamics. The method differs from current methods including adaptive control, variable structure control, and learning control. For this concept, at each sampling time the system response (position, velocity, and acceleration) is observed and used directly to modify controller actions rather than to adjust the controller gains as is the case for the other methods. Preliminary investigations indicate that the proposed concept has promise. The project formulates the fundamentel concept, examines stability, and also the hardware implementation of the controller. The concept is tested on a robot manipulator at the MIT Laboratory for Manufacturing and Productivity.