Scholarlink

Autonomous air combat maneuvering (ACM) is a challenging issue in robotics and artificial intelligence, as ACM structurally exhibits the characteristics of high dimensionality, continuous state, and nonlinear environments. However, the recent advancements in deep reinforcement learning have shown that such aspects can be resolved, enabling unmanned aerial vehicles and similar systems to maneuver and carry out operations effectively during combat. This review paper explores the innovative developments achieved in reinforcement learning techniques to be applied to Close-Air Combat (CAC), focusing on the significant retrieved techniques and outcomes. These approaches incorporate several methods, such as learning algorithms, reward functions, and policies that enhance learning speed. Further, a multi-agent system has been explored to address issues of partial observability in the environment and maintain cooperation among several UAVs. Some suggestions for future research streams are mentioned to eliminate the existing limitations and pave the way for more advancements in intelligent air combat.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

A Brief Review on Reinforcement Learning Based Close Air Combat

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Nahid Anowar,
  • Joydev Debnath,
  • Arup Bhattacharjee,
  • Chandrajit Choudhury

摘要

Autonomous air combat maneuvering (ACM) is a challenging issue in robotics and artificial intelligence, as ACM structurally exhibits the characteristics of high dimensionality, continuous state, and nonlinear environments. However, the recent advancements in deep reinforcement learning have shown that such aspects can be resolved, enabling unmanned aerial vehicles and similar systems to maneuver and carry out operations effectively during combat. This review paper explores the innovative developments achieved in reinforcement learning techniques to be applied to Close-Air Combat (CAC), focusing on the significant retrieved techniques and outcomes. These approaches incorporate several methods, such as learning algorithms, reward functions, and policies that enhance learning speed. Further, a multi-agent system has been explored to address issues of partial observability in the environment and maintain cooperation among several UAVs. Some suggestions for future research streams are mentioned to eliminate the existing limitations and pave the way for more advancements in intelligent air combat.