Real-time sustainable cobotic disassembly planning using fuzzy reinforcement learning

Collaborative robots (cobots) play a vital role in smart manufacturing, particularly in disassembly processes. Human-robot collaboration (HRC) methods simultaneously leverage the complementary capabilities of humans and cobots, offering promising improvements in disassembly processes. A review of the literature reveals that most proposed HRC disassembly planning models do not incorporate sustainable factors, such as consumed energy, human safety, ergonomic risks, and circularity, in the decision-making process. Furthermore, uncertainties inherent in disassembly processes, such as the quality of recovered parts, are not well-addressed in the literature. This paper presents a novel multi-agent fuzzy reinforcement learning (RL) planning model for sustainable HRC disassembly. In addition to cost elements, the developed model involves social and environmental considerations in the real time planning process. By developing a fuzzy-based environment in the RL architecture, the proposed approach aims to effectively model the probabilistic uncertain parameters involved in the problem. Experimental analysis shows that the model presented in this research outperforms a baseline model, applied to the same case study, in terms of convergence time. Furthermore, in terms of qualitative analysis, the proposed model integrates a more extensive set of features into the planning process compared to recent literature.