Adaptive velocity control for UAV boat landing: A neural network and particle swarm optimization approach

Abstract

Achieving precise landing of Unmanned Aerial Vehicles (UAVs) onto moving platforms, such as Autonomous Surface Vehicles (ASVs), is challenging, particularly in GPS-denied environments with dynamic disturbances. Conventional methods often rely on high-level waypoint navigation, extensive manual tuning, and expensive sensors. In this work, we propose an adaptive Proportional-Integral-Derivative (PID) controller optimization using a Neural Network-Particle Swarm Optimization (NN-PSO) algorithm. The algorithm dynamically tunes the PID controller, significantly reducing manual tuning effort, while relying solely on a low-cost camera and altitude sensor. The NN-PSO algorithm allows the UAV to land with an average error of 5 cm on static platforms and 10 cm on moving boats, based on multiple test flights. Our method also increases the maximum landing speed to 80.9% of the UAV’s top flight speed, a considerable improvement over existing systems. Our approach not only optimizes landing precision but also introduces techniques for ensuring soft landings, reducing oscillations, and preventing target misses. These enhancements make the method robust across varying flight altitudes and ASV speeds. Furthermore, this approach is applicable to a variety of GPS-denied scenarios, including rescue missions, package deliveries, and workspace inspections, without requiring costly equipment or extensive parameter tuning. Field experiments confirm the precision and stability of the proposed system, validating its performance in real-world conditions. [Video]