Leveraging large language models for efficient scheduling in Human–Robot collaborative flexible manufacturing systems

With the increasing demand for customized manufacturing, human-robot collaborative (HRC) systems combine human adaptability with robotic precision, offering a promising solution for flexible production. Unfortunately, real-time scheduling remains a significant challenge due to high demand variability, frequent disruptions, and complex task allocation. To address these issues, we propose an evolutionary scheduling framework utilizing a local large language model (LLM). This framework enhances domain-specific understanding by supervising the fine-tuning of the LLM on scheduling data. Additionally, we introduce a population self-evolution mechanism that incorporates individual co-evolution, self-evolution, and collective evolution to improve the generation of heuristic dispatching rules (HDRs). By leveraging the local LLM, our approach generates dynamic HDRs with lower computational overhead, facilitating effective task allocation and sequencing in HRC scenarios while ensuring data privacy. Validated across 54 real-world HRC scenarios, our method achieves a 21.52% average makespan reduction, compared to baseline methods, demonstrating its potential for flexible manufacturing systems.