A reinforcement learning based collaboration framework for autonomous mobile robots

Abstract

Manufacturing has shifted from mass production to mass customisation. The increased product varieties have created significant challenges in the manufacturing process. This demands reconfigurable work cells and manufacturing lines, faster integration time, reusable robotic systems, reduced factory footprint, high-mix and low-volume productions and reduced programming costs. Therefore, an AI based flexible and adaptive robotic control and multi-robot collaboration system is essential to address these challenges and to autonomously react to the environmental and production line changes without human intervention. Autonomous Mobile Robots (AMR), Fig. 1(a), are proposed to address these agile manufacturing challenges. AMRs are devices that can perform tasks and moving through the environment without the need of a predefined path or intervention from human operators. Integration of AMRs with manipulators (robotic arms) and grippers, Fig. 1(b), can support intelligent gripping and placing tasks, e.g., pick up objects, place them on the AMR platform and move the objects to another place, or pick up the objects and places them to different pallets. In realistic industry environments, there are a large number of possible combinations of these AMRs, the robotics arms, grippers and tasks, e.g., a combination of an AMR/ Robotic Arm/ Gripper can be used for different pick and place tasks, and different combinations of AMR/ Robotic Arm/ Gripper can be used for the same pick and place tasks. Training a machine learning model for each of the combinations is time consuming and not adaptable.