Associative memory network controller for robot manipulators with generalized trajectory learning ability

In this paper, we present an associative memory network (AMN) controller for learning and generalization of robot trajectories. The AMN is a special neural network which consists of an input mapping rule called AMR (associative mapping rule) and WUR (weight update rule). The AMR transforms the input variable into several digits which are used for generating memory addresses and WUR corrects the memory weights based on the output computed by summing the memory weights addressed via AMR. In order to use the AMN as an intelligent controller for learning and generalization of robot trajectories, we introduce a learning rule to transform the learned robot trajectory into learned output signal of AMN which drives the controlled robot manipulator. By using this and generalization property of distributed AMN, we construct an AMN controller for robot manipulators and apply to learning of Cartesian (or task) space trajectories. The feasibility and effectiveness of the proposed AMN controller is shown by computer simulation