Scholarlink

This article describes a study on the use of deep reinforcement learning for autonomous drone navigation, focusing on the field of security and monitoring. The real-world application that this article seeks to justify is how this drone will navigate and deal with adversities encountered during its mission. The validation of the results occurs through simulation on a farm, where the drone follows a predefined path. To achieve this, a method called D4PG, with parallelization and distribution capabilities, was applied to handle the learning aspect. The Hydrone is a hybrid quadcopter drone that operates exclusively in the aerial domain, and it was the chosen model employed in the study. The models were trained using both methods in environments of varying complexity, and the results showed that they were able to converge to the expected outcome in most of the proposed challenges. This is a promising study and may enable the application of these methods in various contexts related to robotics and autonomous aerial navigation.

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

Simulation-Based Approaches for Autonomous Security and Monitoring Using Drones

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Álisson H. Kolling,
  • Marcos Gonçalves,
  • Bruno S. Castro,
  • Gustavo Glass,
  • Leonardo F. Pereira,
  • Natã I. Schmitt,
  • Vitoria Lik,
  • Thássio Gomes,
  • Solon Bevilacqua,
  • Anselmo Cukla,
  • Fernando Gamarra,
  • Fernanda Mota

摘要

This article describes a study on the use of deep reinforcement learning for autonomous drone navigation, focusing on the field of security and monitoring. The real-world application that this article seeks to justify is how this drone will navigate and deal with adversities encountered during its mission. The validation of the results occurs through simulation on a farm, where the drone follows a predefined path. To achieve this, a method called D4PG, with parallelization and distribution capabilities, was applied to handle the learning aspect. The Hydrone is a hybrid quadcopter drone that operates exclusively in the aerial domain, and it was the chosen model employed in the study. The models were trained using both methods in environments of varying complexity, and the results showed that they were able to converge to the expected outcome in most of the proposed challenges. This is a promising study and may enable the application of these methods in various contexts related to robotics and autonomous aerial navigation.