Based on a novel idea to harness the onset of complex nonlinear dynamics in information processing or control systems, chaotic dynamics was introduced in recurrent neural network depending on system parameter values, and was implemented into an autonomous roving robot. The robot can catch, by a few sensors, only rough target information with uncertainty, and was designed to solve two-dimensional mazes using adaptive neural dynamics generated by the recurrent neural network, in which four prototype simple motions are embedded as attractors in the state space of neurons. Adaptive switching of system parameter values in the neural network results in various motions depending on environmental situations and enables to solve ill-posed problems. The results of hardware implementation and preliminary experiments show that, in given two-dimensional mazes, the robot can successfully avoid obstacles and reach the target. Therefore, we believe that chaotic dynamics has novel potential capability in complex control by simple rule, and could be useful to practical engineering application mimicking excellent functions observed in biological systems including brain.
