Realistic Physical Simulation and Analysis of Shepherding Algorithms using Unmanned Aerial Vehicles

Abstract

Advancements in UAV technology have offered promising avenues for wildlife management, specifically in the herding of wild animals. However, existing algorithms frequently simulate two-dimensional scenarios with the unrealistic assumption of continuous knowledge of animal positions or involve the use of a scouting UAV in addition to the herding UAV to localize the position of the animals. Addressing this shortcoming, our research introduces a novel herding strategy using a single UAV, integrating a computer vision algorithm in a three-dimensional simulation through the Gazebo platform with Robot Operating System 2 (ROS2) middleware. The UAV, simulated with a PX4 flight controller, detects animals using ArUco markers and uses their real-time positions to update their last known positions.

The performance of our computer-vision-assisted herding algorithm was evaluated in comparison with conventional, position-aware/dual UAV herding strategies. Findings suggest that one of our vision-based strategies exhibits comparable performance to the baseline for smaller populations and loosely packed scenarios, albeit with sporadic herding failures and performance decrement in very tightly packed flocking scenarios and very sparsely distributed flocking scenarios. The proposed algorithm demonstrates potential for future real-world applications, marking a significant stride towards realistic, autonomous wildlife management using UAVs in three-dimensional spaces.