Perception Aware Quadrotor Control: An Experimental Evaluation

Accurate localization of an Uncrewed Aerial Vehicle (UAV) in GPS-denied environments remains a fundamental and unresolved challenge in robotics. Traditional approaches treat state estimation, trajectory planning, and control as separate modules, often assuming perfect state information. However, when visual sensors are used for localization, this assumption often breaks down under challenging conditions, such as poor lighting or textureless environments, resulting in significant drift and unreliable control decisions. Yet, perception performance can be significantly improved if the UAV’s motion is planned with the constraints and capabilities of onboard vision in mind. In this context, Perception-Awareness (PA) emerges as a strategy to adapt a UAV’s maneuvers based on the real-time perception of the environment. In this paper, we implement and evaluate a PA quadrotor controller through an extensive experimental campaign, demonstrating a reduction in odometry drift of up to 70% compared to a non-PA control approach.