onboard wind estimation

Real-time, high-precision 3D wind vector estimation for UAVs under external dependency-free conditions remains a significant challenge. Method: This paper proposes a fully onboard wind estimation algorithm that integrates a disturbance observer (DOB) with thin-plate spline (TPS)-based nonlinear mapping, coupled with a custom-designed, high-sensitivity aerodynamic anemometer structure. This framework enables, for the first time, high-accuracy online estimation of the vertical wind component during dynamic flight. Contribution/Results: An embedded real-time implementation is validated across diverse scenarios—including wind tunnel tests, indoor/outdoor hover, and aggressive maneuvering flight. Experimental results demonstrate root-mean-square errors (RMSEs) of 0.06 m/s (wind tunnel), 0.22 m/s (outdoor hover), and <0.38 m/s (dynamic flight) for horizontal wind speed; <7.3° for wind direction; and <0.17 m/s for vertical wind—surpassing state-of-the-art methods across all metrics.

This work addresses the challenge of autonomous three-dimensional wind field estimation for small unmanned aerial vehicles (UAVs), which is hindered by the limited accuracy of low-cost sensors and critical for energy-efficient flight. The authors propose a method that fuses an aerodynamic model with an extended Kalman filter (EKF) and incorporates an adaptive moving-average estimator (AMAE). Relying solely on standard onboard IMU and GNSS measurements—without requiring additional anemometers or angle-of-attack/sideslip sensors—the approach enables real-time, accurate estimation of both steady-state and time-varying 3D wind fields. Designed to balance computational efficiency and estimation smoothness, the method has been validated through comprehensive simulations and real-flight experiments, demonstrating its suitability for resource-constrained embedded platforms. The study also systematically analyzes the impact of model inaccuracies and real-world uncertainties on estimation performance.

This work addresses the inefficiency of customized airflow generation in multi-fan vertical wind tunnels by proposing an integrated framework that combines a simplified physics-based model with a sample-efficient online learning approach. Leveraging real-time airflow measurements and feedback, the method iteratively refines a coordinated control policy for multiple fans, enabling flexible and stable synthesis of diverse complex velocity profiles—including uniform, Gaussian, and parabolic distributions. Notably, it achieves, for the first time, tailored airflow generation specifically optimized for passive gliding performance, significantly enhancing aerial robot flight capabilities across varying fan configurations.

This study addresses the challenge of degraded navigation robustness and energy efficiency in aerial robots operating in dense urban environments due to the lack of real-time wind field awareness. To overcome this limitation, the authors propose a novel paradigm for real-time local wind field inference that requires no prior environmental information. By fusing onboard LiDAR ranging data with sparse in-situ wind speed measurements, they develop a data-driven model for local wind prediction and, for the first time, integrate it into a receding horizon optimal controller to jointly optimize collision avoidance and energy efficiency. Combining deep learning, fluid dynamics, and optimal control, the method achieves real-time performance on embedded hardware. Simulations demonstrate significantly reduced collision rates and energy consumption, while wind tunnel experiments validate the algorithm’s feasibility and effectiveness on physical platforms.

In GPS- and vision-denied indoor environments, micro-air vehicles (MAVs) lack reliable external references for state estimation. Method: This paper proposes an airflow-based inertial odometry system using hot-wire anemometers. It first decouples rotor downwash and ground-effect interference; designs a gated recurrent unit (GRU) network to robustly estimate relative airspeed from highly perturbed anemometer signals; and constructs a nonlinear observer jointly modeling sensor biases, enabling tightly coupled fusion of hot-wire anemometer, IMU, electronic speed controller (ESC), and barometer measurements. Results: In a 203-second manually piloted randomized flight, position integration drift is only 5.7 m; takeoff and landing velocities are accurately estimated throughout; IMU and barometer biases are calibrated in real time; and airspeed estimation accuracy improves significantly under calm-air conditions.

This work addresses the significant degradation in trajectory tracking performance of small quadrotors operating in the atmospheric boundary layer under strong turbulent winds. To mitigate this, the authors propose a two-stage learning framework: first, an attention-augmented gated recurrent network leverages onboard kinematic and dynamic data to accurately estimate the local three-dimensional wind field, achieving a horizontal wind speed RMSE of 0.40 m/s and a direction error of 3.2°; second, these wind estimates are integrated into a Proximal Policy Optimization (PPO) reinforcement learning controller to enable wind-aware flight control. This approach represents the first integration of learned wind-field perception with reinforcement learning–based control, reducing trajectory tracking error by 48% compared to a non-wind-aware PD controller across wind speeds of 4–12 m/s, outperforming it in all evaluated scenarios, and maintaining stable flight even under out-of-distribution wind conditions of 13–15 m/s, thereby substantially enhancing wind-robustness.

This work proposes an unscented Kalman filter (UKF) approach based on an SE(3) geometric dynamics model for real-time wind velocity estimation on low-cost quadrotor unmanned aerial vehicles, addressing the limitations of conventional extended Kalman filters (EKFs) in highly nonlinear flight scenarios where both accurate wind estimation and precise trajectory tracking are challenging to achieve simultaneously. By integrating the UKF within the SE(3) framework for wind field perception and coupling it with a geometric controller, the method enhances estimation robustness while preserving high-fidelity trajectory tracking. Simulation results demonstrate that, under strong nonlinear conditions such as high winds, the proposed approach significantly improves wind speed estimation accuracy and reduces trajectory deviation compared to EKF, thereby validating its practicality and reliability in complex meteorological environments.

This work addresses the challenge of jointly estimating wind velocity and vehicle state for quadrotor UAVs operating in SE(3) using only onboard sensors. The authors propose a discrete-time dynamics model based on Lie group variational integrators and, for the first time, apply it to wind estimation, thereby avoiding the approximation errors inherent in conventional discretization methods. Building upon this model, they develop a high-precision joint estimation framework on the SE(3) manifold by integrating both extended and unscented Kalman filters (EKF and UKF). Comprehensive simulations and real-world outdoor experiments demonstrate that the proposed discrete SE(3) model, particularly when coupled with the UKF, significantly outperforms continuous-time counterparts and maintains excellent estimation accuracy and trajectory tracking performance even with low-cost sensors.

This study addresses the lack of objective and interpretable criteria for classifying gust load responses by proposing a machine learning approach based on representative samples. Leveraging representation learning and data summarization algorithms, the method extracts a minimal yet highly representative subset from 3,480 experimental datasets of flying-wing models. By integrating similarity metrics with cluster analysis, it establishes an objective classification framework that transcends specific flight conditions. The resulting taxonomy identifies nine fundamental response types, each characterized by distinct transient features that shed light on underlying, shared aerodynamic mechanisms. This interpretable classification not only facilitates expert analysis but also provides a principled basis for guiding subsequent high-fidelity experiments.