Continual Learning Supporting Human-Robot Collaboration

Continual learning, a Machine Learning (ML) approach, entails incre- mental learning and knowledge building over tasks. This study applies continual object detection capability of a human-robot collaborative assembly system facing frequent product rotation. The research is structured into three phases: Initially, hierarchical clustering is employed to establish a parts structure, facilitating the incorporation of new parts and a ML model is pre-trained based on the clustered part. Subsequently, the ML model is used to verify the accuracy of the assem- bly sequence and enable the robot to accurately select parts during the assembly process. Finally, the actual assembly task by operator and robot is carried out. Any unidentified part is associated in real-time with its category and used to con- tinuously update the dataset for training the model. This facilitates ongoing and continual learning.