This paper presents a novel approach to improving quadrotor control by adapting the trajectory generation component rather than changing the controller. The authors argue that this method can enhance trajectory tracking for quadrotor systems experiencing aerodynamic drag forces, common when carrying payloads. They introduce a ‘drag-aware’ planning problem and a tracking cost function, which measures a controller’s ability to follow a reference trajectory. The function acts as a regularizer in trajectory generation and is learned from simulation data. Experiments show that this approach can reduce tracking error by up to 83%, and real-world tests on the Crazyflie hardware platform confirm its effectiveness.
Publication date: 10 Jan 2024
Project Page: https://github.com/hanlizhang/AeroWrenchPlanner
Paper: https://arxiv.org/pdf/2401.04960
