New Quadrotor Landing Framework Uses Adaptive UKF and MPC for Fixed-Time Descent

A team led by Mohammadreza Izadi and including Steven Waslander (University of Toronto) presented a new estimation and control framework for fixed-time dynamic landing of quadrotors. The core innovation is the integration of nonlinear model predictive control (NMPC) with a real-time minimum-jerk trajectory planner that enforces a precise touchdown time, addressing the challenge of landing on moving platforms with consistent timing. To enhance robustness against sensing quality fluctuations—such as GPS dropouts or camera occlusions—the system uses an adaptive unscented Kalman filter (AUKF) that dynamically updates process and measurement noise covariance matrices online. The authors also provide a reference feasibility analysis proving that minimum-jerk references produce bounded thrust and torque commands under standard tracking assumptions.

The framework was validated both in simulation and on hardware. Results showed repeatable landings and significantly improved accuracy in predicting the moving platform's velocity compared to standard extended Kalman filter (EKF) and unscented Kalman filter (UKF) approaches. The paper notes that the adaptive filter enables robust performance even when sensing quality degrades mid-descent. Applications include autonomous drone delivery to moving trucks, shipboard landings, and aerial refueling. The work has been accepted to the Conference on Robots and Vision (CRV 2026) in Vancouver, Canada.