Self-organizing communication in multiagent systems with neural network control

A distributed system for the control of a robot team on the basis of trainable feedforward neural networks is proposed. A developed test model for a system of automatic mine clearance combines neural network control and a multiagent approach, which represents today's paradigm of designing severely complex distributed systems. The key point of the model developed is the use of a neural network for providing self-organizing communication among cooperative robots. The employment of the neural network (which, along with the control of a robot, provides an adaptive data exchange with its partners) makes it possible to implement the adaptability in behavior of individual robots and provide self-organizing behavior of the entire robot team.