Neural-Fly enables rapid learning for agile flight in strong winds

Abstract

Executing safe and precise flight maneuvers in dynamic high-speed winds is important for the ongoing commoditization of uninhabited aerial vehicles (UAVs). However, because the relationship between various wind conditions and its effect on aircraft maneuverability is not well understood, it is challenging to design effective robot controllers using traditional control design methods. We present Neural-Fly, a learning-based approach that allows rapid online adaptation by incorporating pretrained representations through deep learning. Neural-Fly builds on two key observations that aerodynamics in different wind conditions share a common representation and that the wind-specific part lies in a low-dimensional space. To that end, Neural-Fly uses a proposed learning algorithm, domain adversarially invariant meta-learning (DAIML), to learn the shared representation, only using 12 minutes of flight data. With the learned representation as a basis, Neural-Fly then uses a composite adaptation law to update a set of linear coefficients for mixing the basis elements. When evaluated under challenging wind conditions generated with the Caltech Real Weather Wind Tunnel, with wind speeds up to 43.6 kilometers/hour (12.1 meters/second), Neural-Fly achieves precise flight control with substantially smaller tracking error than stateof-the-art nonlinear and adaptive controllers. In addition to strong empirical performance, the exponential stability of Neural-Fly results in robustness guarantees. Last, our control design extrapolates to unseen wind conditions, is shown to be effective for outdoor flights with only onboard sensors, and can transfer across drones with minimal performance degradation.

$\mathcal{L}_1$Quad: $\mathcal{L}_1$ Adaptive Augmentation of Geometric Control for Agile Quadrotors with Performance Guarantees

“…In [16] and [15], the authors use DNNs to predict the uncertainties resulting from ground effect (in takeoff and landing) and aerodynamic interaction forces (of quadrotors flying in close proximity), respectively. In [17], the authors propose to use domain adversarially invariant meta-learning to learn the shared representation of the aerodynamic effects in different wind conditions, which enables fast adaptation by adjusting the linear weights to a specific wind condition. In [7], two neural networks are used to approximate the translational and rotational uncertainties subject to wind disturbance.…”

“…Therefore, ML tools have recently been integrated with conventional control methods to enable more agility and flexibility. Towards this end, ML tools have been frequently used to learn the residual dynamics, which were integrated with adaptive control (linear reference model) [13], [14], nonlinear control [15]- [17], or MPC [18], [19]. Despite the improved performance achieved by the integrated schemes, the data collection and the training process form a significant overhead before the system with ML tools in the loop can be deployed.…”

Progress in symmetry preserving robot perception and control through geometry and learning

“…Future work will implement the trajectory optimization using this geometric control framework proposed by Teng et al (2022b) for robot control. Another interesting research direction is to incorporate learning into this framework ( Shi et al, 2019 ; Li et al, 2022 ; Ma et al, 2022 ; O’Connell et al, 2022 ; Power and Berenson, 2022 ; Rodriguez et al, 2022 ). In addition, the IIG algorithm ( Ghaffari Jadidi et al, 2019 ), combined with an MPC ( Teng et al, 2021a ), can provide an integrated kinodynamic planner that takes the robot stability, control constraints, and the value of information from sensory data into account.…”

FlowDrone: Wind Estimation and Gust Rejection on UAVs Using Fast-Response Hot-Wire Flow Sensors

“…Current multirotor drones typically treat wind as an external disturbance and rely on a feedback controller to perform gust rejection [6], [7], e.g., using techniques from robust and adaptive control [8]- [10]. Recently, techniques from adaptive control have also been combined with deep neural network representations of aerodynamic effects in order to perform multirotor control in the presence of wind [11].…”