Dynamic task planning for autonomous reconfigurable manufacturing systems by knowledge-based multi-agent reinforcement learning.

Reconfigurable manufacturing systems can rely on the collaboration of autonomous modular machines, dynamically planning production tasks and effectively satisfying demands. This paper introduces a decentralized knowledge-based framework, considering task uncertainties, module specialization, and reconfiguration constraints. By taking advantage of scenario reproducibility of simulations and utilizing deep reinforcement learning with multi-objective rewards, the proposed method enables machines to autonomously make intelligent decisions and collaborate on sequential, co-executed, or parallel tasks. Experiments are designed to evaluate the effectiveness of autonomous machines in planning tasks and the adaptability to uncertain situations in manufacturing operations, showcasing efficient module usage without compromising completion speed. [ABSTRACT FROM AUTHOR]