Motion controlling in 2-D space using chaotic dynamics in a neural network with large redundancy

A controlling method of a moving object in 2-D space is proposed, in which chaotic dynamics introduced into a neural network model is applied to solving 2-D maze, that is one of ill-posed problems. According to simply defined motion functions calculated from the network state at time t, the object moves from time t to t+1. As prototype motions, four original attractors are embedded in our neural network model. They corresponds to the simple motions of the object toward four directions in 2-D space. Introducing chaotic dynamics into the network gives us outputs which sample intermediate state points between embedded attractors in network state space, which enables the object to move toward various directions. By system parameter switching between chaotic regime and stable regime, the object is able to success rate of this method for 300 trials is higher than that of random pattern generator.