Warehouse Robot Task Scheduling Based on Reinforcement Learning to Maximize Operational Efficiency

摘要

This paper focuses on optimizing the task scheduling of warehouse robots through reinforcement learning to improve operational efficiency. With the development of robotics technology, there are still limitations in its intelligence level, and cloud computing and machine learning provide new opportunities for this. The centralized task allocation model of traditional robot systems faces the problem of high computational pressure. The cloud robot architecture transfers some functions to the cloud to relieve the burden. This paper proposes a robot task scheduling strategy based on parallel reinforcement learning, which decomposes complex tasks into sub-problems, utilizes the parallel computing of the cloud platform, and multiple agents asynchronously feedback the learning results. The Q-Learning algorithm is selected to solve the sub-problems, the convergence of its asynchronous parallel computation is analyzed, and a knowledge transfer method is constructed. The experiment is carried out on a computing cluster built with two Inspur servers, and the software is extended based on CloudSim. The results show that this strategy is superior to the random and round-robin scheduling strategies under each number of computing nodes. When the number of computing nodes is small, increasing the number of nodes can improve the performance, while when the number is large, the improvement is limited. It provides technical support for the development of intelligent logistics systems.