Cooperative battle method and device for intelligent agents

Abstract

The present invention relates to a method, device, terminal equipment and storage medium for cooperative combat of intelligent agents. By determining the virtual air combat scene where the intelligent agents fight; according to the virtual air combat scene, the action space information and state space of one or more intelligent agents are determined. Information, according to the state value, determine the reward value of the action corresponding to the state value; according to the virtual air combat scene, action space information, state space information and reward value, train the initial reinforcement learning model, when the initial reinforcement learning model is in a convergent state , get the target reinforcement learning model; use the target reinforcement learning model to fight against the regular agent, overcome the difficulty and instability of the target reward, and when the strategy of the multi-agent is changed, it will not affect the reinforcement learning of the multi-agent.

Description

technical field [0001] The present invention relates to the technical field of artificial intelligence, in particular to a method, device, terminal equipment and storage medium for cooperative combat of intelligent agents. Background technique [0002] When multiple unmanned combat aircraft make autonomous maneuver decisions in air combat, they need to carry out decision-making cognition and coordination. Because the environment of unmanned combat aircraft is relatively complex, and the coordination between aircraft needs to be considered, how to realize the autonomous control of combat aircraft is a research focus. [0003] Traditional UAV control relies on expert knowledge to deal with different situations through expert judgment on the environment and experience construction rules, but this requires experts to have high experience and knowledge, and it takes a lot of time and energy to consider all situations. With the development of artificial intelligence technology, d...

AI Technical Summary

Problems solved by technology

However, in multi-agent reinforcement learning, the environment is complex and changeable, the state space will grow exponentially with the increase of agents, the problem of dimension explosion may occur, and the calculation is complicated; at the same time, there are difficulties and instability in determining the target reward. The definition of the reward function will be affected by the different cooperation and tasks between multi-agents, and when the strategy of each agent changes, the strategies of other agents will also change, affecting the final convergence of the algorithm

Images