Robust and Agile Quadrotor Flight via Adaptive Unwinding-Free Quaternion Sliding Mode Control

--This paper presents a new adaptive sliding mode control (SMC) framework for quadrotors that achieves robust and agile flight under tight computational constraints. The proposed controller addresses key limitations of prior SMC formulations, including (i) the slow convergence and almost-global stability of SO(3)-based methods, (ii) the oversimplification of rotational dynamics in Euler-based controllers, (iii) the unwinding phenomenon in quaternion-based formulations, and (iv) the gain overgrowth problem in adaptive SMC schemes. Our controller is computationally efficient and runs reliably on a resource-constrained nano quadrotor, achieving 250 Hz and 500 Hz refresh rates for position and attitude control, respectively. In an extensive set of hardware experiments with over 130 flight trials, the proposed controller consistently outperforms three benchmark methods, demonstrating superior trajectory tracking accuracy and robustness with relatively low control effort. The controller enables aggressive maneuvers such as dynamic throw launches, flip maneuvers, and accelerations exceeding 3g, which is remarkable for a 32-gram nano quadrotor . The experimental codes and videos related to this paper are accessible at the following links: Code: https://github.com/A A. Motivation Quadrotors require robust control to maintain stability and precise maneuverability under disturbances and uncertainties. One widely studied method in this context is sliding mode control (SMC). One key challenge involves attitude control. As discussed in Section II, coordinate-free methods exhibit slow convergence and provide only almost global stability. The authors are with the Autonomous V ehicles Laboratory, Department of Aerospace Engineering, Toronto Metropolitan University, Toronto, Canada{amin.yazdanshenas,reza.faieghi Quaternion-based methods also face the unwinding issue, which can cause unnecessarily prolonged rotations. A second challenge is the need to know the upper bounds of uncertainties. Adaptive switching gains eliminate the need for prior knowledge of these bounds.