Data-Driven Reinforcement Learning for Optimal Motor Control in Washing Machines

In this paper, we address the challenge of developing advanced motor control systems for modern washing machines, which are required to operate under various conditions. Traditional system designs often rely on manual trial-and-error methods, limiting the potential for performance enhancement. To overcome this, we propose a novel continual offline reinforcement learning framework, specifically tailored to improve balance maintenance during the dehydration cycle of washing machines. Our approach introduces a delayed online update mechanism that leverages accumulated transition data from certain periods of online interaction. This method effectively circumvents the distribution shift problem commonly encountered in offline reinforcement learning. Our empirical results demonstrate a substantial improvement, with an average increase of nearly 16% in load balancing efficiency across various tasks, including those involving different types of laundry. This research not only enhances the applicability of reinforcement learning in industrial settings but also represents a significant step forward in the development of smart appliance technology.