Optimization of Multi-Agent Scheduling Based on MA-ID3QN for the Riveting and Welding Work Cell

In industrial manufacturing, multi-agent scheduling is one of the key technologies for improving production efficiency. Due to the complexity of multi-agent systems and the interference between tasks, achieving efficient task scheduling is faced with significant challenges. To solve this problem, this paper introduces the dueling double deep Q-network (D3QN) into the multi-robot scheduling scenario of a riveting and welding work cell for the first time. Considering the characteristics of this scenario, an improved D3QN is proposed, which is designed as a multi-agent independent dueling double deep Q-network algorithm (MA-ID3QN) based on a multi-agent cooperation mechanism. In this approach, robots in the work cell are treated as independent agents, with decentralized training and decentralized execution to accommodate varying robot numbers. Meanwhile, several mechanisms are employed to enhance the algorithm’s performance. Furthermore, a digital twin-based riveting and welding work cell platform is constructed for validation. First, the MA-ID3QN algorithm generates a scheduling strategy based on the state of the physical space of the riveting and welding work cell. Then, the scheduling strategy is verified on the digital twin platform. Finally, comparative experiments are conducted to validate the effectiveness of the proposed method. The experimental results demonstrate that the MA-ID3QN-based agent scheduling method exhibits better reliability, higher efficiency, and stronger generalization capability in multi-agent task scheduling. This approach improves the efficiency of the riveting and welding work cell and reduces the time required for welding tasks in mass production scenarios. Moreover, it has promising application prospects in industrial robot scheduling.