Hierarchical structure of deterministic chaos in a chaotic neural network model (CNN) is investigated in the view of application in robotics. The result shows a rich capacity of CNN in selectively controlling the synchronization of neuron outputs, and sensitively responding to external sensory inputs, both being based on the intrinsic mechanism of the dynamics called chaotic itinerancy. Choosing appropriate parameters, the simple designed robot realized a chaotic search to the hierarchically selected directions. The macroscopic drift preserving chaotic fluctuation was also derived by simply adding weak external inputs to an intended direction. Obstacle avoidance was simulated with the use of these properties.