Cloud robot task scheduling method and system based on parallel reinforcement learning

Abstract

The embodiment of the invention provides a cloud robot task scheduling method and system based on parallel reinforcement learning. The cloud robot task scheduling method includes the steps that original problems are loaded in a scheduling center, and the original problems are divided into a plurality of sub-problems; the sub-problems and a plurality of computational nodes are matched by the scheduling center; the sub-problems are sent to the computational nodes matched with the sub-problems by the scheduling center; the computational nodes carry out parallel reinforcement learning on the sub-problems and feed learning results of the sub-problems back to the scheduling center; and the scheduling center judges whether the original problems have converged or not according to the learning results of the sub-problems and preset convergence conditions, and if the original problems have converged, optimal strategies of the original problems are output. The method can obtain the precise optimal strategies and reduce the time cost.

Description

technical field [0001] The present invention relates to a cloud robot task scheduling strategy based on parallel reinforcement learning, which belongs to the field of machine learning, relates to the combination and use of cloud robots, distributed computing, and reinforcement learning, and specifically relates to a cloud robot task scheduling based on parallel reinforcement learning methods and systems. Background technique [0002] In recent years, robots have entered a period of rapid development, and the rise in labor costs has given rise to the need to use machines to replace manpower. At present, because the capabilities of robots, especially the level of intelligence and expectations, are far apart, the application of commercial robots is mainly concentrated in the fields of large-scale repetitive production such as automobiles and electronic equipment. With the widespread use of cloud computing, whether it is renting public clouds or deploying local clouds, it provi...

AI Technical Summary

Problems solved by technology

This kind of episodic tasks can be solved by table solving algorithm, but as the state space expands, the time required to complete the learning increases rapidly, and it is unacceptable for too much time overhead in practical applications; and the approximate solution method uses the finite state space Effective promotion of experience, when encountering unknown situations, generalize similar situations from previously encountered situations. The key lies in the generalization of the problem, and it is difficult to obtain the optimal strategy for the overall problem.

Images