Deep Learning Methods for Decentralized Decision-Making in Counterswarm Engagements is an open-source Naval Postgraduate School teza by Kurdo A. Sharif.
The rise of unmanned technologies has fueled interdisciplinary research into robotic swarm systems, particularly within military applications. These systems draw inspiration from the problem-solving capabilities of biological swarms, offering the benefit of emergent global behavior that arises from local interactions and minimizes the need for centralized control. Traditional methods for creating emergent behavior in robotic swarms depend on predictable and controllable swarm dynamics, clearly defined local rules, and complete knowledge of all agents.
In contrast, counter-swarm strategies require robust and adaptable global approaches that function effectively in dynamic environments with limited information. This research explores the inverse problem: designing local rules to achieve emergent behaviors typically derived from perfect knowledge and communication among drones. The goal is to develop decentralized regions where a defender drone employs a neural network model trained extensively on simulation data.
The simulation data, derived from scenarios involving three attackers and one defender, was organized into various input sets representing different features. After training, regression analysis was performed to identify which feature set produced the most effective defender heading angles compared to an oracle algorithm. The findings revealed that the neural network model outperformed the oracle in optimizing shorter engagements, demonstrating the potential of using trained networks as an alternative to traditional algorithms.
Data publikacji– June 2024
Deep Learning Methods for Decentralized Decision-Making in Counterswarm Engagements contains the following major sections:
- Wprowadzenie
- Generating Simulation Data
- Training the Defender
- Neural Network Performance and Post-Training Analysis
- Wnioski
Źródło zdjęcia - Adobe Stock by Constantine Art (Generated with AI)