Countering a Drone in a 3D Space: Analyzing Deep Reinforcement Learning Methods is a work by Ender Çetin, Cristina Barrado, and Enric Pastor.
In the industry, numerous methods exist for countering drones. However, employing an artificial intelligence (AI)-based counter-drone system can prove to be an efficient approach, eliminating the need for human intervention. This paper introduces a novel method using deep reinforcement learning (DRL) to counter drones in a three-dimensional (3D) space by utilizing another drone. While the effectiveness of deep reinforcement learning in countering drones has already been demonstrated in a two-dimensional (2D) space, tackling the challenge in a 3D environment presents additional complexities, such as training time and obstacle avoidance. To address these challenges, a Deep Q-Network (DQN) algorithm incorporating a dueling network architecture and prioritized experience replay is proposed. The algorithm aims to capture another drone within an Airsim simulator-based environment. The models are trained and tested using diverse scenarios to evaluate the drone’s learning progress. To enhance training, the experiences gained from previous training sessions are transferred by preprocessing and eliminating unfavorable experiences. The results highlight the significant improvement achieved through transfer learning, leading to enhanced drone learning progress. Furthermore, an algorithm combining imitation learning and reinforcement learning, referred to as deep Q-learning from demonstrations (DQfD), is implemented to capture the target drone. This algorithm samples expert demonstrations data along with self-generated data by the agent, allowing the agent to continue learning without overwriting the demonstration data. The primary advantage of this algorithm is its ability to expedite the learning process, even with limited demonstration data available.
Publication Date: November 2022
Countering a Drone in a 3D Space: Analyzing Deep Reinforcement Learning Methods contains the following major sections:
- Contributions and Related Work
- Materials and Methods
- Training and Test Results
All articles published by MDPI are made immediately available worldwide under an open-access license. No special permission is required to reuse all or part of the article published by MDPI, including figures and tables. For articles published under an open-access Creative Common CC BY license, any part of the article may be reused without permission, provided that the original article is clearly cited. For more information, please refer to https://www.mdpi.com/openaccess.
C-UAS Hub does not own this content and provides a link for users at the bottom of the page to access it in its original location. This allows the author(s) to track important metrics related to their work. All credit goes to its rightful owner.
Authors:Ender Çetin, Cristina Barrado, and Enric Pastor
For additional multimedia resources, please visit the Multimedia Library.
Check out suppliers of drone detection and mitigation systems at C-UAS Hub Products.
Stay on top of industry news, developments, resources and articles- Sign up for a free C-UAS Hub Membership to bookmark your favorite content and receive the weekly newsletter and important industry updates!