Evolution of shape-changing and self-repairing control for the ATRON self-reconfigurable robot

Abstract

The ATRON self-reconfigurable robot consists of simple interconnected modules. Modules move relative to other modules and as a result change the shape of the robot. The ATRON modules are difficult to control because of complex motion constraints on the modules. Motion constraints are reduced by using meta-modules composed of three modules. A meta-module may emerge from unstructured groups of modules if three modules are connected in the right configuration. The meta-module then moves on a surface of modules and stop at another position. To attract moving meta-modules and thereby to specify the shape-changing task of the robot we use attraction-points. In this work we evolve a distributed artificial neural network controller for the modules. The controller is identical on every module and controls when a meta-module emerges, how it move and when it stops. In simulation we demonstrate how this control strategy allows the ATRON robot to shape-change to support an unstable roof, build a bridge across a gap and to self-repair a broken bone. We conclude that the control strategy is able to shape-change and self-repair the ATRON robot independent on whether it consists of dozens, hundreds or thousands of modules