Digital Twin-Based Transfer Learning for Collaborative Robot Systems: A Proof of Concept

Nowadays, the increasing trend toward digitalization has driven the extensive adoption of collaborative robotic automation across industries, yet a significant limitation is the robots' adaptability to unexpected and dynamic environments. This research introduces a Digital Twin (DT)-based Transfer Learning (TL) approach that combines DTs and Machine Learning (ML) to enhance adaptability in collaborative robot systems. The proposed system uses DT cyberspace for pre-training ML algorithms and leverages TL to apply this knowledge to real-world applications. This innovative approach efficiently trains state-of-the-art ML models, delivering exceptional performance while reducing the required time and data resources. The proof-of-concept experiments, employing the proposed DT-based TL to control soccer robots, demonstrate a remarkable 96% reduction in training time while maintaining a high level of adaptability, achieving a 70% goal accuracy rate in dynamic scenarios.