Knowledge sharing-enabled low-code program for collaborative robots in mix-model assembly

Abstract

Multi-robot collaboration is a crucial execution tool for mixed-model assembly lines. The rapid reconfiguration of the robots with impaired skills to maintain the robustness of the assembly line remains a significant challenge. With a focus on knowledge-driven faster transition technologies for collaborative robots, this paper proposes a Knowledge Sharing-enabled Low-code Program (KSLC) method to address the deficient skill migration and the limited scalability caused by programs written statically in open-loop control. First, considering collaborative robots' functional requirements and environmental constraints, the parameterized action primitive library of assembly skills is developed with properties across multiple perspectives, levels, and granularities. Complex assembly skills are then formally expressed using the Web Ontology Language (OWL). Besides, digraph network model is created to represent action sequences and the corresponding parameters relevant to complex assembly tasks for the execution content. Finally, the DQN algorithm is utilized to learn low-dimensional vectors within the knowledge graph. The GraphSAGE algorithm is employed to facilitate skill search and matching, enabling the effective acquisition and transmission of robot skills. Experimental results demonstrate that the proposed KSLC-enabled collaborative robots achieve 90 % average success rate in the TwoArmPegInHole task, significantly outperforming the traditional experience transfer strategies that only attain 58 % success rate. This finding indicates that KSLC can substantially enhance robot learning efficiency and task performance.