Learning-based multi-robot collaboration for intelligent and adaptive robotic systems

The concepts of Artificial Intelligence (AI) and Digital Twins are playing a more and more important role in industrial manufacturing, which can be adopted to empower more collaborative and adaptive multi-robot systems to replace humans in finishing some repetitive tasks. Existing studies are more focused on static environments or solutions that rely on centralized coordination. This thesis carried out two case studies as part of the academic initiative called "Conscious Agents as a Foundation for a Collaborative Factory of the Future.". The initial study presents an environment for a manufacturing scenario, and an observation algorithm that records the activity of an autonomous robot executing pick-and-place operations and merges this knowledge with reinforcement learning for motor control and decision support. The subsequent study proposes a method of decentralized learning in a dynamic multi-robot collaborative system. Tests confirm that the suggested learned policy allows robots to avoid collisions while working with another controller by the same policy, and also ensures the performance of independent work. This work contributes to the understanding of how decentralized learning can enhance autonomy and cooperation in industrial robotic systems.