Task agent implementation method based on finite state machine and deep reinforcement learning

By dividing tasks into multiple execution phases and using finite state machines and deep reinforcement learning to train the agent, the problems of rationality and simplicity in task execution in complex combat scenarios are solved, and a high level of intelligent autonomous control is achieved.

CN118333088BActive Publication Date: 2026-06-26HANGZHOU EBOYLAMP ELECTRONICS CO LTD
View PDF 2 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
HANGZHOU EBOYLAMP ELECTRONICS CO LTD
Filing Date
2024-03-15
Publication Date
2026-06-26

AI Technical Summary

Technical Problem

Existing technologies make it difficult to build task agents capable of intelligent and autonomous control in complex combat scenarios, especially in multi-stage situations where it is difficult to achieve rational and simplified task execution.

Method used

By employing a method based on finite state machines and deep reinforcement learning, the task is divided into multiple execution stages, and transition conditions between stages are constructed. The action space, situation space, neural network structure, and reward function of the agent are configured, and the agent's autonomous decision-making is achieved through training.

Benefits of technology

It enables the scientific organization of complex tasks, ensuring the rationality and simplicity of task execution, improving the reusability of modules, and training high-level intelligent agents that exceed human capabilities.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118333088B_ABST
    Figure CN118333088B_ABST
Patent Text Reader

Abstract

The application discloses a task intelligent agent implementation method based on a finite state machine and deep reinforcement learning, which is applied to a two-party confrontation process, and the two parties each include a preset number of formations, each formation has a task to perform, and the two-party confrontation process is as follows: the task to perform is divided into a preset number of execution stages, and a conversion condition between the execution stages is constructed; the execution stage is taken as a state of a finite state machine, and the finite state machine is executed according to the conversion condition between the execution stages; an intelligent agent is set for each execution stage, and an action space of each intelligent agent is configured; each intelligent agent is trained based on the finite state machine and the action space; the task to perform is implemented in stages by using the finite state machine technology, scientific process organization and stage behavior execution can be performed on a complex execution task containing multiple action sequences, the rationality, simplicity and expansibility of the task execution process are ensured, and the reusability of the module is improved.
Need to check novelty before this filing date? Find Prior Art