Self-organizing CPGs in the control loop of a biomorphic fish robot

Abstract

The rhythmic motor activity of animals is based on the work of interconnected populations of neurons, the so-called central pattern generators (CPGs). Recently, the use of artificial CPGs as controllers of robotic devices, in particular in fish robots, has been proposed. In the current paper, we present a software package that allows simulating CPG and demonstrating their work using a virtual robot. Using the developed tools, we have studied the possibility of CPG self-organization based on the formation of closed neural circuits in a network with Spike-timing-dependent plasticity (STDP). Modeling activity in modular neural networks with STDP revealed an increase in unidirectional intermodular connections leading to circulating dynamics. In the case of bidirectional intermodular connections, network exhibits a variety of dynamic patterns of neural activity. The use of a self-organizing CPG in the control loop of a virtual fish robot showed the gradual formation of a locomotor pattern as the neural network is structurally and functionally reorganized.