<p>With the rapid development of drone technology, the autonomous collaborative control of drone swarms has been extensively applied across various domains. However, in some cluttered environments, drone swarms need to perform rescue missions without centralized control, relying on onboard controllers instead, which brings challenges for the collaboration and obstacle avoidance of drone swarms. In this paper, we propose a flocking control model for drone swarms, Convolutional Soft Actor-Critic (CSAC), based on deep reinforcement learning (DRL). Specifically, we integrates DRL with convolutional neural networks (CNNs), aiming to accelerate the end-to-end decision-making process by reducing model parameters. With this framework, each drone can independently make decisions based on its current state and environmental information, ensuring the autonomous execution of collaborative rescue missions in cluttered environments. Through simulation experiments, we have validated the effectiveness of the model. Besides, compared to the baseline algorithm, our model improves decision performance by 5.90<InlineEquation ID="IEq1"> <InlineMediaObject> <ImageObject Color="BlackWhite" FileRef="10586_2025_5475_Article_IEq1.gif" Format="GIF" Height="16" Rendition="HTML" Resolution="72" Type="Linedraw" Width="15" /> </InlineMediaObject> <EquationSource Format="TEX">\(\%\)</EquationSource> <EquationSource Format="MATHML"><math> <mo>%</mo> </math></EquationSource> </InlineEquation> while reducing the number of parameters by 92.91<InlineEquation ID="IEq2"> <InlineMediaObject> <ImageObject Color="BlackWhite" FileRef="10586_2025_5475_Article_IEq1.gif" Format="GIF" Height="16" Rendition="HTML" Resolution="72" Type="Linedraw" Width="15" /> </InlineMediaObject> <EquationSource Format="TEX">\(\%\)</EquationSource> <EquationSource Format="MATHML"><math> <mo>%</mo> </math></EquationSource> </InlineEquation>.</p>

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

Autonomous collaborative rescue of drone swarms in cluttered environments

  • Xiangyu Zhao,
  • Zipeng Zhao,
  • Yu Wan,
  • Jun Tang

摘要

With the rapid development of drone technology, the autonomous collaborative control of drone swarms has been extensively applied across various domains. However, in some cluttered environments, drone swarms need to perform rescue missions without centralized control, relying on onboard controllers instead, which brings challenges for the collaboration and obstacle avoidance of drone swarms. In this paper, we propose a flocking control model for drone swarms, Convolutional Soft Actor-Critic (CSAC), based on deep reinforcement learning (DRL). Specifically, we integrates DRL with convolutional neural networks (CNNs), aiming to accelerate the end-to-end decision-making process by reducing model parameters. With this framework, each drone can independently make decisions based on its current state and environmental information, ensuring the autonomous execution of collaborative rescue missions in cluttered environments. Through simulation experiments, we have validated the effectiveness of the model. Besides, compared to the baseline algorithm, our model improves decision performance by 5.90 \(\%\) % while reducing the number of parameters by 92.91 \(\%\) % .