UAV flight control sensing enhancement with a data-driven adaptive fusion model

Accurate sensing is essential for achieving reliable control of unmanned aerial vehicles (UAVs). In prior works, the unscented Kalman filter (UKF) has shown superior performance in both estimation accuracy and computation efficiency, which makes it suitable for onboard sensing enhancement. However, the prediction accuracy of existing filter-based statistical models is generally assumed to be invariant of the flight conditions. To avoid deterioration in the estimation performance, this paper proposes a novel data-driven adaptive fusion state space model for quantifying the prediction uncertainty of the system model. Based on the multi-output Gaussian process regression (GPR), a rule for tuning the noise parameter of the statistical model is provided based on the estimated variance. The sparse GPR model is utilized to incorporate available features and obtain high estimation accuracy under dynamic operation conditions. Simulation results have illustrated the superior performance of the proposed approach in state estimation for the UAV.