Reinforcement Learning Algorithms for Autonomous Mission Accomplishment by Unmanned Aerial Vehicles: A Comparative...

Area of Science:

• Robotics and Control Systems
• Artificial Intelligence
• Machine Learning

Background:

• Unmanned Aerial Vehicles (UAVs) are increasingly utilized in diverse applications.
• Artificial Intelligence (AI), particularly Reinforcement Learning (RL), offers advanced control methods for UAVs.
• Efficient UAV navigation requires balancing speed, distance, and obstacle avoidance.

Purpose of the Study:

• To compare the performance of three RL algorithms: Deep Q-Network (DQN), SARSA, and Advantage Actor-Critic (A2C).
• To evaluate these algorithms for UAV navigation from point A to point B in a neighborhood environment.
• To assess the algorithms' ability to minimize time and distance while avoiding obstacles.

Main Methods:

• Implementation and comparison of DQN, SARSA, and A2C algorithms.
• Simulation of UAV navigation in a representative neighborhood environment.
• Performance evaluation based on successful navigation, time, and distance metrics.

Main Results:

• DQN successfully achieved the navigation target.
• SARSA and A2C did not achieve the target under the initial fixed premises.
• Analysis indicated potential for A2C to outperform DQN with fine-tuning, showing faster convergence.

Conclusions:

• DQN demonstrated robust performance for UAV navigation in the tested environment.
• SARSA and A2C require further optimization for comparable performance.
• A2C presents a promising alternative due to its simpler structure and potential for faster learning with adjustments.