The main objective of this research is to explore the development of an intelligent memory that can be trained to remember the previous control commands for various trajectories of a robot manipulator. In contrast to the conventional analytical approach, this technique will provide a real-time, learning, and adaptive control which does not require any accurate models of the system. In initial work, a simple three-degree-of-freedom manipulator resembling the Stanford arm was used as a model to generate a thousand quantified trajectory points for a training set, and a simple averaging interpolation formula was used to estimate joint commands for other trajectories. A hierarchical (tree) structure was explored for storing the configuration space joint data for the training set.