A Goal-Directed Trajectory Planning Using Active Inference in UAV-Assisted Wireless Networks.

Area of Science:

• Robotics
• Wireless Communications
• Artificial Intelligence

Background:

• Unmanned aerial vehicles (UAVs) offer flexible aerial base station deployment to augment terrestrial networks.
• Effective flight trajectory planning is crucial for maximizing the benefits of UAV-assisted wireless communications.
• Current methods may lack the adaptability and efficiency needed for dynamic UAV communication scenarios.

Purpose of the Study:

• To introduce a novel, goal-directed trajectory planning method for UAVs using active inference.
• To enhance wireless connectivity between UAVs and terrestrial users.
• To develop a robust and efficient path planning solution for UAV base stations.

Main Methods:

• A global dictionary representing a world model was created using traveling salesman problem with profits (TSPWP) instances.
• The world model uses 'letters' (hotspots), 'tokens' (local paths), and 'words' (trajectories) to encode decision-making grammar.
• Active inference enables the UAV to interpret the world model and deduce optimal routes based on belief states.

Main Results:

• The proposed active inference method demonstrated superior performance compared to traditional Q-learning.
• The method provides fast, stable, and reliable solutions for UAV trajectory planning.
• The approach exhibits good generalization ability across various scenarios.

Conclusions:

• Active inference offers a powerful framework for intelligent trajectory planning in UAV-assisted wireless networks.
• The developed world model facilitates efficient route optimization and decision-making for UAVs.
• This approach significantly improves wireless connectivity and network reinforcement capabilities.