<p>Industrial robots are widely used in automotive spot-welding lines to realize large-scale flexible production, where multiple spot-welding robots (SWR) require collaboration to complete a series of welding tasks. While traditional genetic algorithm-based approaches have been widely used for path planning and optimization in robotics, it still faces limitations in handling the challenges of multiple spot-welding robots in a dynamic production environment. Therefore, this paper proposes a novel reinforced pointer network for path-planning solution of multiple spot-welding robots. First, a digital twin-based co-optimization architecture is proposed for task assignment and path planning of multiple SWR. Subsequently, a novel reinforced pointer network is constructed by the actor-critic model for path planning of multiple SWR, where a mathematical model is designed to minimize the maximum completion time as the primary objective function. Furthermore, k-means and genetic algorithms are employed to enhance the efficient allocation of tasks in multi-robot. Finally, the effectiveness of the proposed method is verified by applying it to the welding tasks and path planning of the rear assembly, in comparison to the traditional bilayer chromosome coding method. The results show that the proposed method can acquire better results after only a few generations, considerably improving search efficiency and responding to realize autonomous dynamic path planning under abnormal conditions. In the practical application of the automobile production line, the bottleneck station is optimized from 189.5 to 184.5&#xa0;s.</p>