Abstract
In this paper, we estimate the continuous-time nonlinear rotational dynamics of a DJI F450 quadcopter which is controlled and stabilized by an onboard unrooted Android phone with a PID controller. Since the Android phone contains the only IMU in the system, we only have measurements provided by the Android Java API for estimation, which are subject to significant latencies because Android is not realtime system. Although prediction error methods are typically preferred for estimating general continuous-time nonlinear models, we show that the standard prediction error estimates are highly sensitive to the initial guess, and hence we introduce a total least-squares approach for estimating the initial model. Finally, we compare the estimated models to the model obtained from a traditional inertia measurement device, namely, a Bifilar test stand.