Constrained Kalman Filter for Adaptive Prediction in Minidrone Flight

Abstract

The minidrone Parrot Mambo® is a promising robotic platform for education control purposes. An important limitation is that its SDK provides sensor data with a maximum nominal frequency of just 2 Hz, creating objective difficulties for feedback control. This paper proposes an observer capable of generating prediction on the data, which allows feeding the controller with a much faster rate than the one allowed by the slow sensor data rate. The predictions are generated by a linear model, whose parameters are identified on-line using a Constrained Kalman Filter. The strategy is successfully validated via extensive experiments with real drones performing altitude stabilisation and trajectory tracking tasks. In particular, the constrained model identification preserves a stable prediction (which is physically meaningful), and hence safe flight, even in the presence of large disturbances.