On the Effectiveness of Torque Sensor on Cooperative Transport for Robot Swarm

Cooperative transport in swarm robotics is particularly challenging because the group coordination needs to be generated without using centralized control, global information and any form of direct communication. This study investigates a simple scenario for cooperative object transport in robotics swarm. The aim is to investigate the effects of torque sensors on generating effective group transport strategies for a robot swarm. The torque sensor is simulated by reading the torque feedback from Bullet physical engine hinge constraint which connects every wheel to the body of the robot. In this simulated study, a group of two agents controlled by a dynamic neural network shaped by artificial evolution are required to transport an elongate cuboid object which cannot move by single agent. We examine the flexibility of best evolved controller against different classes of object mass and size. The evolved controller exhibits a re-orientation, re-position and communicating forces through the object behaviors analogous to those observed in a biological ant. In spite of task complexity, the results demonstrate a decent performance in term of adaptability and flexibility to different object's size, mass and to different starting conditions.