Adaptive SINDy: Residual Force System Identification Based UAV Disturbance Rejection

A research team including Fawad Mehboob, Amir Atef Habel, and Dzmitry Tsetserukou has published a paper on arXiv introducing Adaptive SINDy, a novel AI-driven control system designed to help drones reject wind disturbances. The system uniquely integrates two established techniques: Sparse Identification of Nonlinear Dynamics (SINDy), a data-driven method for discovering the governing equations of a system, and Recursive Least Square (RLS) adaptive control. This hybrid approach aims to solve the persistent challenge of keeping Unmanned Aerial Vehicles (UAVs) stable in highly nonlinear, turbulent environments where traditional analytical modeling falls short and purely learning-based methods lack interpretability.

The team validated Adaptive SINDy in both Gazebo simulations and real-world flights using a very lightweight Crazyflie drone. They created a highly dynamic test environment with wind speeds of up to 2 meters per second blowing from four different directions. The AI controller was tasked with tracking complex circular and lemniscate (figure-eight) trajectories. The results were significant: Adaptive SINDy outperformed conventional baseline controllers like PID and Incremental Nonlinear Dynamic Inversion (INDI) on several error metrics, all while avoiding crashes. It achieved a root mean square error (RMSE) as low as 12.2 cm on a circular path and 17.6 cm on a lemniscate, with corresponding mean absolute errors (MAE) of 13.7 cm and 10.5 cm, demonstrating precise and robust disturbance rejection.