[Paper Review] Precision Landing of a UAV on a Moving Platform for Outdoor Applications

Adarsh Salagame, Sushant Govindraj+1 more|arXiv (Cornell University)|Sep 28, 2022

Robotics and Sensor-Based Localization4 citations

TL;DR

This paper presents a vision-based precision landing system for a fixed-wing UAV on a moving outdoor platform using monocular camera tracking of Apriltag bundles and PID control. The system achieves stable tracking and successful landing at speeds up to 1.5 m/s under windy pre-monsoon conditions, demonstrating robustness to outdoor disturbances like wind and variable lighting.

ABSTRACT

As UAV technology improves, more uses have been found for these versatile autonomous vehicles, from surveillance to aerial photography, to package delivery, and each of these applications poses unique challenges. This paper implements a solution for one such challenge: To land on a moving target. This problem has been addressed before with varying degrees of success, however, most implementations focus on indoor applications. Outdoor poses greater challenges in the form of variables such as wind and lighting, and outdoor drones are heavier and more susceptible to inertial effects. Our approach is purely vision based, using a monocular camera and fiducial markers to localize the drone and a PID control to follow and land on the platform.

Motivation & Objective

• To enable autonomous precision landing of a UAV on a moving outdoor platform despite environmental challenges like wind and variable lighting.
• To address the limitations of prior indoor-focused or sensor-dependent approaches by using a purely vision-based method with fiducial markers.
• To ensure continuous target tracking during descent, even when field of view shrinks, by employing multiple-sized Apriltag bundles.
• To validate the system's robustness under real-world outdoor conditions with dynamic target motion and environmental disturbances.
• To demonstrate feasibility of autonomous landing on a moving platform using a heavier outdoor drone (3DR Iris, 2.5kg) with onboard processing.

Proposed method

• A monocular downward-facing camera on a 3DR Iris drone captures images of Apriltag bundles on a moving platform for real-time localization.
• Apriltag bundles—groups of markers in decreasing sizes—enable continuous detection from 5m down to landing, mitigating field-of-view limitations.
• The system computes position errors (∆x, ∆y) between the drone and target, which are fed into a PID controller to generate velocity commands.
• The landing process is staged: tracking initiates at 4m altitude, and landing begins after 2 seconds of continuous detection, with altitude increase if detection is lost.
• A dynamic recovery strategy raises the drone to increase field of view and reacquire the target if tracking is lost.
• The system uses an Odroid XU4 for onboard image processing and wireless telemetry to a ground station for monitoring.

Experimental results

Research questions

• RQ1 Can a purely vision-based system achieve reliable tracking and precision landing of a UAV on a moving outdoor platform under variable wind and lighting conditions?
• RQ2 How does the use of multiple-sized Apriltag bundles improve detection reliability during descent compared to single markers?
• RQ3 What is the impact of PID gain selection on tracking efficiency and system stability when the target moves at varying speeds?
• RQ4 How does the system recover from temporary loss of target detection during flight?
• RQ5 To what extent can the system maintain stable tracking and landing performance on a moving platform with a heavier outdoor UAV?

Key findings

• The system achieved a tracking efficiency of 93.35% at a platform speed of 0.5 m/s with optimized PID gains (kp=0.65, ki=0.0015, kd=0.09).
• Tracking efficiency decreased significantly with increasing platform speed, dropping to 23.36% at 1.0 m/s with lower gains, indicating sensitivity to target dynamics.
• Higher PID gains (e.g., kd=0.09) led to instability and oscillations at higher speeds, causing loss of tracking despite initial detection.
• The system successfully landed the drone on a moving rover traveling at 1.5 m/s during outdoor flight tests, confirming real-world viability.
• When tracking was lost, the drone autonomously ascended to increase field of view and successfully reacquired the target, demonstrating robust recovery behavior.
• The algorithm stabilized within approximately 25 seconds of continuous tracking, showing convergence to steady-state flight above the moving target.