Adaptive Control of Robot Systems with Simple Rules Using Chaotic Dynamics in Quasi-layered Recurrent Neural Networks

Abstract

A novel idea of adaptive control with simple rules using chaotic dynamics in a recurrent neural network model and two kinds of quasi-layered recurrent neural network model have been proposed. Since chaos in brain was discovered in the context of brain function, the authors have claimed that chaos has complex functional potentialities and have presented the results of computer experiments which use chaos to solve several kinds of “ill-posed problems”. The key idea is to harness the onset of complex nonlinear dynamics in dynamical systems. More specifically, attractor dynamics and chaotic dynamics in a recurrent neural network model are introduced by changing a system parameter, “connectivity” in one type of model and via “sensitive response of chaos to external inputs” in other models. In this report, we will show the following. (1) A global outline of our idea and our recurrent neural network models with neuro-chaos , (2) Several computer experiments on the use of the neuro-chaos recurrent neural network models for solving of 2-dimensional mazes by an autonomous robot, in the context of an ill-posed problem setting, (3) Hardware implementations of the computer experiments using robots with two-wheels or two-legs driven by a neuro chaos simulator. Successful results of maze-solving are shown not only for computer experiments but also for practical experiments, (4) A proposal for a pseudo-neuron device using semiconductor and opto-electronic technologies. The device is called a “dynamic self-electro optical effect devices (DSEED)”, and it has the potential to be a “neuromorphic device” or even a “brainmorphic device”. (5) A proto-type model of intra-brain communications between far distant neurons in the brain is proposed, from a heuristic point of view based on observations of neuron synchronization phenomena associated with advanced brain functioning.KeywordsComplex dynamicsChaosAdaptive controlQuasi-layered recurrent neural networkHardware implementationAutonoumous robotNeuromorphic deviceBrainmorphic device