(series: Drones, Data and Errors in Agricultural Environments)
Drift observed during monitoring flights is not an isolated failure, but a consequence of inaccuracies in the onboard state estimation. The drone’s flight control system continuously determines position, orientation, and motion state, thereby maintaining an internal representation of the aircraft’s state. The construction and maintenance of this state representation rely on data continuously provided by multiple, independent sensor systems.
The GNSS system supplies information related to the drone’s geographic position and velocity. However, these data cannot be considered absolutely accurate: GNSS always provides an environment-dependent estimate whose precision may vary both temporally and spatially. During monitoring flights, GNSS data often appear stable, while minor inaccuracies may already be present without immediately indicating a critical deviation. In many cases, such deviations are not readily detectable by the average operator.
This external position estimate is complemented by the inertial measurement unit (IMU), which measures linear accelerations, angular rates, and changes in orientation. The advantage of the IMU lies in its ability to track motion rapidly and with high precision; however, over longer time periods and in the absence of external reference, accumulated errors may occur. This phenomenon becomes particularly relevant during monitoring operations in which the drone remains in a steady flight regime for extended durations.
The onboard flight control system performs state estimation by fusing data from GNSS, the IMU, and onboard optical sensors. Flight log analysis from both examined monitoring operations revealed time intervals during which individual sensor sources provided differing position or motion parameters. During these periods, control authority was maintained; however, uncertainty in position estimation increased. This manifested as a gradual lateral deviation between the flight path computed by the onboard controller and the planned mission trajectory.
Monitoring flights typically involve flight profiles with limited dynamic maneuvers, which supports stable control but simultaneously reduces the amount of feedback available to the system. In such conditions, minor discrepancies between sensor inputs do not trigger immediate corrective action, but instead become gradually incorporated into the onboard state model. Initially, this process does not result in visible anomalies; over time, however, it may lead to the emergence of drift.
In the next part, a concrete case study will illustrate how this deviation develops into an operational flight event in practice, and which indicators were already identifiable in the onboard data prior to the complete loss of control.