Application of Deep Reinforcement Learning to UAV Swarming for Ground Surveillance

Area of Science:

• Artificial Intelligence
• Robotics
• Aerospace Engineering

Background:

• Current aerial swarm management relies on classical and reinforcement learning methods.
• There is a need for advanced AI systems for surveillance and law enforcement applications.

Purpose of the Study:

• To propose a hybrid AI system integrating deep reinforcement learning into a centralized swarm architecture.
• To enable aerial swarms for effective area surveillance, ground target search, and tracking.

Main Methods:

• A hybrid AI system with a central swarm controller and deep reinforcement learning (DRL) trained agents.
• Utilized Proximal Policy Optimization (PPO) algorithms for agent behavior training.
• Defined metrics to assess swarm performance in surveillance and tracking tasks.

Main Results:

• Simulations show the system effectively searches operational areas.
• The system demonstrates efficient target acquisition within reasonable timeframes.
• Consistent and continuous tracking of acquired targets was achieved.

Conclusions:

• The proposed hybrid AI system offers an effective solution for aerial swarm-based surveillance.
• Deep reinforcement learning enhances the capabilities of cooperative UAVs for security applications.
• The system provides reliable performance in searching, acquiring, and tracking ground targets.