Adaptive Control of Modular Robots

This paper proposes a learning control system of modular systems with a large number of degrees of freedom based on joint learning of modules, starting with finding the common control rules for all modules and finishing with their subsequent specification in accordance with the ideas of the semantic probabilistic inference. With an interactive 3D simulator, a number of successful experiments were carried out to train three robot models: snake-like robot, multiped robot and trunk-like robot. Pilot studies have shown that the approach proposed is quite effective and can be used to control the complex modular systems with many degrees of freedom.