WAND: Learning Robust Navigation under Complex Wind Disturbances and Dense Obstacles for Quadrotors

Zhonghan Tang; Chenhui Li; Shuai Liang; Zhongrui You; Jianan Li; Bin Zhao; Zhigang Wang; Xuelong Li

cs.ROOct 8, 2026

WAND: Learning Robust Navigation under Complex Wind Disturbances and Dense Obstacles for Quadrotors

Authors: Zhonghan Tang, Chenhui Li, Shuai Liang, Zhongrui You, Jianan Li, Bin Zhao, Zhigang Wang, Xuelong Li

Organizations: Department of Electronic Engineering and Information Science, University of Science and Technology of China, Hefei 230027, China · Shanghai Artificial Intelligence Laboratory, Shanghai 200232, China · Northwestern Polytechnical University, Xi’an 710072, China · Institute of Artificial Intelligence (TeleAI), China Telecom, Beijing 100033, China

Abstract

Robust navigation in cluttered environments remains a fundamental challenge for quadrotors, particularly when strong wind disturbances arise, which perturb vehicle dynamics, limit control authority, and substantially increase collision risk. Existing learning-based navigation policies typically rely on obstacle perception and proprioceptive observations, requiring the policy to infer time-varying disturbance effects implicitly and thereby limiting robustness under partial observability. This paper proposes WAND (Wind-Aware Navigation with Disturbance Estimation), a reinforcement learning framework for navigation under time-varying wind disturbances in dense obstacle fields. Specifically, WAND estimates wind-induced disturbance acceleration from historical proprioceptive states using a Temporal Convolutional Network (TCN). This estimation is integrated into the policy via a zero-initialized residual module, \emph{WindAdapter}, while simultaneously providing feedforward compensation for low-level control. The dual use of the estimate couples disturbance-conditioned navigation with feedforward disturbance rejection. Across 12 wind-disturbed simulation settings, WAND improved the observed success rate by 8.3 percentage points on average relative to feedforward compensation alone. Controlled opposite-crosswind experiments further showed wind-direction-dependent trajectory adaptation. In indoor fan-induced flight tests, WAND succeeded in 18 of 20 trials, demonstrating the feasibility of real-time onboard navigation.

Figures & tables

Fig. 1 : Real-world deployment of WAND. (a) A quadrotor navigates a cluttered indoor scene under fan-induced airflow; the inset shows an anemometer reading of approximately $7.2\,\mathrm{m/s}$ . (b) The corresponding visualization in RViz.

Fig. 2 : Overview of WAND framework. Ray-cast obstacle perception is first encoded into compact features, while a TCN-based wind disturbance estimator predicts the wind-induced disturbance acceleration from a temporal history of proprioceptive states. The estimated disturbance is encoded by WindAdapter and fused with the state feature to form the input to the policy network. The policy outputs a desired acceleration command, which is further combined with the feedforward disturbance compensation term for execution.

Fig. 3 : Analytical wind-velocity profiles used in simulation.

Fig. 4 : Reach-goal rates during PPO training: (a) navigation-only comparison and (b) WAND ablations. Curves show the mean over five independent random seeds, and shaded regions indicate 95% confidence intervals across seeds.

Fig. 5 : Navigation success rates under constant, turbulent, gust, and mixed wind. Markers show success rates over 50 trials; error bars indicate 95% Wilson confidence intervals. Markers are slightly offset horizontally for clarity.

Method	Constant	Turbulent	Gust	Mixed
EKF	0.3204	0.4620	0.4203	0.4474
LSTM	0.1166	0.2405	0.2059	0.2525
TCN	0.0777	0.1876	0.1159	0.1970

TABLE I : RMSE(

m/s^{2}

) IN DIFFERENT WIND TYPES

Panel of Figure 7 — Fig. 6 : Fixed-scene trajectories under opposite $7.5\,\mathrm{m/s}$ crosswinds. The translucent volume and arrows indicate the prescribed wind region and direction. Each panel overlays 20 trials per method; WAND and Base+Comp are shown in blue and orange, while red crosses and yellow squares mark collisions and timeouts.

Wind	Method	S/C/T (%)	$\bm{d_{\min}}$ (m)
$+\mathcal{W}x$	Base+Comp	60/20/20	$0.086\pm 0.089$
	WAND	100/0/0	$\mathbf{0.376\pm 0.022}$
$-\mathcal{W}x$	Base+Comp	0/100/0	–
	WAND	100/0/0	$\mathbf{0.449\pm 0.107}$

TABLE II : FIXED-SCENE RESULTS UNDER OPPOSITE CROSSWINDS.

Fig. 7 : Real-world trajectories under (a) weak and strong fixed-fan wind and (b) strong handheld-fan wind.

Method	Scene 1	Scene 2	Total
EGO-Planner	4/10	6/10	10/20
NavRL	5/10	6/10	11/20
Base+Comp	7/10	6/10	13/20
WAND	10/10	8/10	18/20

TABLE III : REAL-WORLD NAVIGATION RESULTS (SUCCESSES/TRIALS)

Explore similar work

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
Anticipatory Risk-Guided Reinforcement Learning for Safe Flight Through Dynamic Clutter
Jul 26, 2026Yuchao Mei, Guohao Zhang, Luxia Ai +2Collision Avoidance Risk-Sensitive RL
National Key Laboratory of Science and Technology on Multi-spectral Information Processing, School of Artificial Intelligence and Automation, Huazhong University of Science and Technology, Wuhan 430074, China
Adaptive Outer-Loop Control of Quadrotors via Reinforcement Learning
May 15, 2026Vishnu Saj, Sushil Vemuri, Dileep Kalathil +1RL Control Sim-to-Real Transfer
Texas A&M University College Station, TX, USA

Jul 1, 2026, cs.LGJ/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

Explore similar work

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

Anticipatory Risk-Guided Reinforcement Learning for Safe Flight Through Dynamic Clutter

Adaptive Outer-Loop Control of Quadrotors via Reinforcement Learning