Method for reconfiguring the kinematic structure of a mechatronic-modular robot in non-deterministic conditions

Abstract

Modular robots, consisting of many identical modules, are one of the most difficult areas of robotics. Each newly added element changes the shape and capabilities of the end device, for example, adds functionality or allows the robot to move in new planes. The reconfiguration of the kinematic structure is a sequence of movements of each robot module from the initial position of the initial configuration to the final position of the desired configuration. The paper considers a method for reconfiguration the kinematic structure of a mechatronic-modular robot using reinforcement learning. The proposed method will be built on the basis of a learning algorithm, where the information for training will be the actions taken and the "reward" is a value characterizing the quality of the robot's completion of the target task. The purpose of the training is to build a control algorithm that maximizes the total reward for a certain period of time. The effectiveness of the learning algorithm was tested by computer simulation of a robot, consisting of 5, 10 and 15 modules, in the formation of the target configuration