Multi-agent training method and system under complex conditions

Abstract

The invention provides a multi-agent training method and system under complex conditions, and the method comprises the steps: building a training scene and an agent model based on a training target; sequentially decomposing the training scene into a plurality of course tasks from simple to complex according to the complexity of the scene; sequentially selecting curriculum tasks for training according to the scene complexity by utilizing the agent model to obtain strategies of each agent; wherein the training result of the previous course task is used as the initial condition of the next coursetask in the training process. According to the invention, the training scene is sequentially decomposed into a plurality of course tasks from simple to complex according to the complexity of the scene, the problem of instability in multi-agent reinforcement learning is solved, the training effect is improved, and the training time is shortened at the same time.

Description

technical field [0001] The invention relates to the technical field of agent control, in particular to a training method and system for multi-agents under complex conditions. Background technique [0002] In the study of multi-agent systems, an intuitive research method is to predefine behavior rules for agents. In task execution, each agent implements various behaviors according to preset rules. However, this method needs to define a large number of behavioral rules to deal with various possible situations in the environment. In a complex environment, it is difficult to enumerate various situations in the environment, and the behavior of other agents will make the environment change continuously. Therefore, in a complex environment, multi-agents need to learn new behaviors through continuous interaction with the environment to ensure the execution performance of tasks. Reinforcement learning (RL), as a learning mode that does not rely on prior knowledge and data, is an e...

AI Technical Summary

Problems solved by technology

[0005] In summary, in the training process of multi-agent deep reinforcement learning with complex scenarios, especially in the cooperation / competition scenario of training multi-agents, with the increase of agents, the existing training methods have training instability, Poor training effect and long training period

Images