Delay-Informed Intelligent Formation Control for UAV-Assisted IoT Application.

Area of Science:

• Robotics and Control Systems
• Artificial Intelligence
• Internet of Things (IoT)

Background:

• Unmanned aerial vehicles (UAVs) are increasingly vital for IoT applications, with formation flying enhancing surveillance and security.
• The leader-follower approach simplifies UAV formation control by focusing on the leader's trajectory.
• Dynamic leader velocities and communication time delays pose significant challenges for real-time UAV formation control.

Purpose of the Study:

• To investigate the design of UAV formation tracking using deep reinforcement learning (DRL) in high-mobility scenarios with communication delays.
• To formulate the UAV formation problem as a state error minimization problem considering communication delays.
• To develop a DRL algorithm capable of compensating for performance degradation caused by time delays.

Main Methods:

• Formulation of the UAV formation problem as a state error minimization using a quadratic cost function.
• Development of a delay-informed Markov decision process (DIMDP) incorporating previous actions to mitigate time delay effects.
• Proposal of an extended-delay informed deep deterministic policy gradient (DIDDPG) algorithm.

Main Results:

• The proposed DIDDPG algorithm effectively compensates for performance degradation caused by communication time delays.
• Numerical experiments validated the algorithm's ability to maintain stable formation tracking despite delays.
• The study analyzed computational complexity and the impact of varying time delays on system performance.

Conclusions:

• Deep reinforcement learning offers a robust solution for UAV formation control in the presence of communication delays.
• The developed DIDDPG algorithm provides a significant advancement in real-time UAV formation tracking for high-mobility scenarios.
• The approach is extendable to arbitrary communication delay scenarios, broadening its applicability.