Vehicle-State-Aware Confidence Estimation for Direct Wind Sensing on a Multirotor

Sri Arjun Senthil Kumar; April Zi Qian Chan; Chen-Hua Yeow

cs.ROOct 19, 2026

Vehicle-State-Aware Confidence Estimation for Direct Wind Sensing on a Multirotor

Authors: Sri Arjun Senthil Kumar, April Zi Qian Chan, Chen-Hua Yeow

Abstract

Direct wind sensing from a multi-rotor UAV enables localized atmospheric measurements, but the reliability of the anemometer depends on vehicle motion. This work presents a vehicle-state-aware confidence method that uses horizontal and vertical vehicle speed to quantify the acquisition of each wind observation. The resulting confidence score is used to adapt measurement influence to a two-dimensional vector estimator. Experiments were conducted during horizontal translation, vertical motion, and held-out autonomous waypoint flight. The proposed two-feature formulation achieved an ROC-AUC of 0.995, with 98.84% of translation observations identified as unreliable. The confidence-aware adaptation reduced median estimator gain during translation from 0.207 to 0.064 while retaining median gain of 0.151 during hover. The results show that vehicle state provides useful proprioceptive context for regulating the influence of direct airborne wind observations.

Figures & tables

Fig. 1: Representative D6 MeteoNode platform and onboard sensing architecture. (a) Outdoor experimental platform equipped with an elevated HY-WDC2E ultrasonic anemometer. (b) Environmental and vehicle-state measurements are acquired through the Raspberry Pi companion computer and Pixhawk flight controller and recorded through ROS 2 for downstream confidence-aware wind processing.

Fig. 2: Characterization of the representative A6 ultrasonic anemometer and its interaction with the multirotor platform. (a) Validated first-order wind-speed calibration relation over the experimentally evaluated range. (b) Wind-direction validation at eight reference orientations; error bars denote circular standard deviation. (c) Change in measured wind speed under propulsion for three external-flow conditions, showing that elevated mounting does not completely eliminate vehicle-induced aerodynamic influence.

Fig. 3: Vehicle-state-aware confidence formulation for direct multirotor wind sensing. Calibrated wind observations and synchronized vehicle state are temporally associated to determine measurement availability. The direct wind branch forms the navigation-frame observation and calibration-derived nominal covariance $R_{0,k}$ , while horizontal and absolute vertical vehicle speeds are normalized using the frozen stable-hover reference to obtain the continuous confidence $R_{w}$ . The resulting confidence scales the measurement covariance before the wind-vector estimator update.

Model	Horizontal AUC	Vertical AUC
$M_{1}:\{v_{h}\}$	0.980	0.639
$M_{2}:\{v_{h},v_{z,\mathrm{abs}}\}$	0.975	0.866
$M_{3}:+\theta_{\mathrm{tilt}}$	0.972	0.851
$M_{4}:+\omega$	0.973	0.857

TABLE I: Vehicle-state feature ablation.

Panel of Figure 5 — Fig. 4: Held-out AUTO-flight validation using the frozen vehicle-state-aware wind-confidence formulation. (a) Spatial visualization of direct wind observations along the repeated A–B waypoint trajectory. The gray line denotes the recorded UAV trajectory; arrow orientation represents the measured navigation-frame wind-vector direction, arrow length is proportional to wind speed, and arrow color denotes confidence $R_{w}$ . Every eighth synchronized observation is displayed for clarity, with no spatial interpolation. (b) Horizontal and absolute vertical vehicle-speed features relative to the frozen stable-hover upper references; shaded regions denote translation phases. (c) Continuous wind-observation confidence $R_{w}$ , with qualitative confidence thresholds and the same independently defined translation intervals. (d) Median effective Kalman gain for the fixed-covariance and confidence-aware estimators across the corresponding flight phases.

Evaluation	Hover median $R_{w}$	Motion median $R_{w}$	ROC–AUC	Motion classified Unreliable (%)
Horizontal development	0.981	$\approx 0$	0.975	91.4
Vertical validation	0.947	$\approx 0$	0.866	69.7
Held-out AUTO	0.797	$3.31\times 10^{-21}$	0.995	98.84

TABLE II: Summary of vehicle-state-aware confidence validation.

Confidence Estimation

See all 42 papers

Wind-Aware Reinforcement Learning Control of a Small Quadrotor Using Learned Onboard Wind Estimation in Simulated Atmospheric Turbulence
Jul 1, 2026Abdullah Al Tasim, Wei SunRL Control Aerial Robotics
University of Oklahoma, Norman, Oklahoma, 73019
Onboard Wind Estimation for Small UAVs Equipped with Low-Cost Sensors: An Aerodynamic Model-Integrated Filtering Approach
Apr 22, 2026Bingchen Cheng, Tielin Ma, Jingcheng Fu +2UAV Navigation Extended Kalman Filtering
Shenyuan Honors College, Beihang University, Beijing, 100191, China · Institute of Unmanned System, Beihang University, Beijing 100191 · School of Aeronautical Science and Engineering, Beihang University, Beijing 100191, China
Towards Agile Vision-Based Multi-UAV Flight: Revisiting State Estimation
Sep 30, 2026Michal Pliska, Matouš Vrba, Ondřej Víta +3Aerial Robotics Kalman Filtering
Multi-robot Systems Group, Faculty of Electrical Engineering, Czech Technical University in Prague, Czech Republic

Jul 1, 2026J/K move · Enter open · S save

Wind-Aware Reinforcement Learning Control of a Small Quadrotor Using Learned Onboard Wind Estimation in Simulated Atmospheric Turbulence

Abdullah Al Tasim, Wei Sun

University of Oklahoma, Norman, Oklahoma, 73019

RL Control Aerial Robotics Quadrotor Control

Open paper AskPDF arXiv

Small multirotor aircraft are increasingly tasked with operations in the atmospheric boundary layer, where turbulent winds comparable to the vehicle's airspeed degrade trajectory tracking and can defeat conventional feedback control. This work illustrates a two-stage learning pipeline that first estimates the local wind from onboard kinematics and dynamics and then exploits that estimate inside a reinforcement learning (RL) flight controller. The wind estimator, an attention-augmented gated recurrent network trained on thousands of simulated flights through von Karman turbulence with power-law shear and veer, recovers the horizontal wind vector with a per-flight root-mean-square error of 0.40 m/s and a direction error of 3.2 degrees on unseen wind regimes, an accuracy near the floor imposed by unresolved turbulence, and generalizes to vertical ascent profiles with a skill score of 0.861 over a constant-wind reference. A proximal policy optimization controller receiving the frozen estimator's output reduces horizontal trajectory tracking error by 48% relative to a wind-blind proportional-derivative baseline across mean winds of 4 m/s to 12 m/s, winning on 100% of evaluation episodes. A three-way ablation decomposes this improvement into a kinematic component, available without wind information, and a wind-perception component; the perception share rises with wind speed, from small in light winds toward roughly half the total benefit in strong winds, consistent with the quadratic scaling of aerodynamic drag. The controller degrades gracefully on out-of-distribution winds of 13 m/s to 15 m/s, where the baseline fails catastrophically.

Abstract

Confidence Estimation

Explore similar work

Wind-Aware Reinforcement Learning Control of a Small Quadrotor Using Learned Onboard Wind Estimation in Simulated Atmospheric Turbulence

Onboard Wind Estimation for Small UAVs Equipped with Low-Cost Sensors: An Aerodynamic Model-Integrated Filtering Approach

Towards Agile Vision-Based Multi-UAV Flight: Revisiting State Estimation