Visual-Attention-Based Neighbor Selection for Artificial-Potential-Field UAV Formation Control

Area of Science:

• Robotics
• Artificial Intelligence
• Aerospace Engineering

Background:

• Unmanned Aerial Vehicle (UAV) formation technology is rapidly advancing.
• Neighbor selection for UAVs without reliable radio communication presents challenges due to redundant state processing in traditional methods.
• Complex conditions necessitate efficient and robust neighbor selection strategies.

Purpose of the Study:

• To design and evaluate novel visual-attention-based neighbor-selection algorithms for UAVs.
• To address the limitations of fixed-radius and fixed-topology methods in complex formation flying scenarios.
• To improve the efficiency and effectiveness of neighbor selection in decentralized UAV systems.

Main Methods:

• Development of two algorithms: Visual Attention Potential Field (VAPF) and Cluster Visual Attention Potential Field (CVAPF).
• Utilizing a zoom-lens visual attention rule to identify informative neighbors before control input evaluation.
• Implementing a dual-layer communication architecture in CVAPF for cluster-level selection and improved synchronization.

Main Results:

• VAPF and CVAPF significantly reduce the number of processed neighbors compared to existing methods (e.g., BOIDS, Optimized-flocking) in redundant sensing scenarios.
• VAPF and CVAPF maintain essential formation behavior while processing fewer neighbors (VAPF: 6.31, CVAPF: 3.60 on average).
• CVAPF demonstrates improved synchronization and gathering capabilities in clustered formations.

Conclusions:

• Visual-attention-based neighbor selection offers a more efficient approach for UAV formations.
• VAPF and CVAPF provide a robust solution for decentralized UAVs operating without reliable radio communication.
• These algorithms enhance scalability and performance in complex, multi-UAV systems.