UAV Autonomous Tracking and Landing Based on Deep Reinforcement Learning Strategy

Area of Science:

• Robotics and Control Systems
• Artificial Intelligence
• Aerospace Engineering

Background:

• Unmanned aerial vehicles (UAVs) require advanced autonomous capabilities for tracking and landing, especially in unpredictable environments.
• Machine learning, particularly deep reinforcement learning, offers promising solutions for complex robotic control tasks.

Purpose of the Study:

• To develop a novel, model-free approach for UAV autonomous tracking and landing using deep reinforcement learning.
• To address challenges posed by harsh environments, sensor noise, and intermittent measurements in UAV operations.

Main Methods:

• A partially observable Markov decision process (POMDP) framework was employed, utilizing an end-to-end neural network.
• The approach combines the Deep Deterministic Policy Gradients (DDPG) algorithm with heuristic rules for learning landing maneuvers.
• A Modular Open Robots Simulation Engine (MORSE)-based reinforcement learning framework was utilized for validation.

Main Results:

• The proposed deep reinforcement learning algorithm demonstrated an average landing success rate approximately 10% higher than the Proportional-Integral-Derivative (PID) method.
• The UAV successfully learned optimal landing strategies in simulations involving a randomly moving platform with high sensor noise and intermittent data.

Conclusions:

• The study validates a state-of-the-art deep reinforcement learning-based UAV control method for autonomous landing.
• The developed approach offers improved performance and robustness for UAVs operating in challenging and dynamic conditions.