FlowDrone Wind Estimation and Gust Rejection on UA Vs Using Fast-Response Hot-Wire Flow Sensors Nathaniel Simon Allen Z. Ren Alexander Piqu e David Snyder Daphne Barretto

2025-04-27 1 0 4.19MB 7 页 10玖币

侵权投诉

FlowDrone: Wind Estimation and Gust Rejection on UAVs

Using Fast-Response Hot-Wire Flow Sensors

Nathaniel Simon, Allen Z. Ren, Alexander Piqu´

e, David Snyder, Daphne Barretto,

Marcus Hultmark, and Anirudha Majumdar

Abstract— Unmanned aerial vehicles (UAVs) are ﬁnding use

in applications that place increasing emphasis on robustness

to external disturbances including extreme wind. However,

traditional multirotor UAV platforms do not directly sense wind;

conventional ﬂow sensors are too slow, insensitive, or bulky

for widespread integration on UAVs. Instead, drones typically

observe the effects of wind indirectly through accumulated

errors in position or trajectory tracking. In this work, we

integrate a novel ﬂow sensor based on micro-electro-mechanical

systems (MEMS) hot-wire technology developed in our prior

work [1] onto a multirotor UAV for wind estimation. These

sensors are omnidirectional, lightweight, fast, and accurate.

In order to achieve superior tracking performance in windy

conditions, we train a ‘wind-aware’ residual-based controller

via reinforcement learning using simulated wind gusts and

their aerodynamic effects on the drone. In extensive hardware

experiments, we demonstrate the wind-aware controller out-

performing two strong ‘wind-unaware’ baseline controllers in

challenging windy conditions. See: youtu.be/KWqkH9Z-338.

I. INTRODUCTION

Autonomous multirotor drones have the potential to trans-

formatively impact a variety of domains including infra-

structure inspection and repair, search-and-rescue operations,

and aerial package delivery. Towards this end, a major

challenge facing current systems is their limited ability to

deal with severe wind conditions in outdoor environments,

which in turn may impair reliable operations for each of

the applications listed above. For example, the maximum

safe wind speed for a typical multirotor drone (e.g., the DJI

Phantom [2]) is around 20 miles per hour, which roughly

corresponds to a windy day at the beach. Further, the

challenge posed by wind is exacerbated by the presence

of complex airﬂow phenomena (e.g., ground and surface

effects) when the drone operates in proximity to obstacles

or in urban canopies.

Currently, multirotor systems rely almost exclusively on

indirect estimates of wind for real-time control, e.g., on

deviations from nominal motion measured by an inertial

N Simon, A Z Ren, A Piqu´

e, D Snyder, M Hultmark, and A

Majumdar are with the Department of Mechanical and Aerospace

Engineering, and D Barretto is with the Department of Computer Science,

Princeton University, Princeton, NJ, 08544. Emails: {nsimon,

allen.ren, apique, dasnyder, daphnegb,

hultmark, ani.majumdar}@princeton.edu

This work was partially funded by the Air Force Ofﬁce of Scientiﬁc

Research [FA9550-22-1-0020], the National Science Foundation GRFP

[DGE-2039656], and Princeton University’s Project X Innovation Fund. Any

opinions, ﬁndings, and conclusions or recommendations expressed in this

material are those of the author(s) and do not necessarily reﬂect the views

of the AFOSR, NSF, or Princeton. M. Hultmark is co-founder and CEO of

Tendo Technologies, Inc., who provided the MEMS hot-wire dies.

Fig. 1: FlowDrone incorporates the MAST (top right), an omnidirectional

ﬂow sensor leveraging MEMS hot-wires to provide fast and accurate estim-

ates of wind magnitude and direction, enabling superior ﬂight performance

in windy conditions.

measurement unit (IMU), GPS, or downward-facing optical

ﬂow camera. These estimates rely on gusts impacting the

drone’s motion before they can be reliably estimated, and

are hence inherently limited in terms of their accuracy and

update rates. Existing sensors for direct wind measurement

(e.g., conventional pitot tubes and hot-wires) are typically

either too slow, insensitive, or lack the requisite form-factor

(e.g., size and weight), and are thus rarely deployed on

multirotor systems (see Sec. I-A for an overview of existing

wind sensing technology).

The primary hypothesis behind our work is that advance-

ments in airﬂow sensing technology will enable signiﬁcant

improvements in multirotor drone performance in severe

wind conditions. In this paper, we propose the use of novel

micro-electro-mechanical systems (MEMS) technology for

turbulent airﬂow measurements [3], [4]. These sensors have

three characteristics that make them ideally suited for use

in multirotor control. First, they are small (on the order of

a few millimeters) and lightweight (on the order of a few

grams). Second, they have very low latency (<2ms) [1]

and can operate at rates commensurate with typical drone

control rates (>500Hz). Third, they allow accurate real-

time estimation of wind magnitude and direction [1]. As

such, we believe that MEMS-based airﬂow sensors can ﬁll an

important gap in sensing technology for multirotor drones.

Statement of contributions. The primary contribution of

arXiv:2210.05857v2 [cs.RO] 25 Oct 2022

this work is to leverage MEMS technology for turbulent

airﬂow measurements in order to improve multirotor drone

performance in high winds. To this end, we make the

following three speciﬁc contributions:

•FlowDrone. We instantiate the MEMS-based ﬂow sens-

ing technology in the FlowDrone (Fig. 1): a quadrotor

equipped with MEMS hot-wire airﬂow sensors, which

enable real-time (>500Hz) estimates of wind magnitude

and direction.

•Wind-aware control. We propose a reinforcement learn-

ing (RL) pipeline for training a wind-aware controller

that utilizes the wind estimate from the ﬂow sensor. The

key components of the RL training pipeline are: (i) a

simulator that captures relevant aerodynamic effects, (ii)

the simulation of wind gusts recorded from real experi-

ments with additional domain randomization, and (iii) a

neural network policy that takes as input a history of wind

estimates along with the drone’s state estimate and outputs

a residual control input added to the open-source PX4

controller [5].

•Hardware experiments. We perform hardware experi-

ments demonstrating the improved performance of the

FlowDrone’s wind-aware controller in the presence of

a wind gust as compared to two strong wind-unaware

baselines: (i) the widely-used PX4 controller (referred to

as ‘baseline’), and (ii) a controller trained using RL with

identical simulated winds but without the ability to access

real-time wind estimates (referred to as ‘wind-unaware’).

Over the course of 30 ﬂights (10 per controller), each

exposed to a 5 m/s gust, the mean maximum error in the

direction of wind was: 0.44 m for wind-aware, 0.58 for

wind-unaware, and 0.78 for baseline. This constitutes a

44% improvement of wind-aware over baseline.

A. Related Work

Multirotor drone control in wind. 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].

Alternative approaches to wind estimation include utilizing

an extended Kalman ﬁlter with measurements from the

drone’s IMU (e.g., as implemented by the widely used PX4

controller [12]). Such wind estimates can then be utilized by

a feedback controller for gust rejection [13].

The approaches mentioned above rely on indirect es-

timates of wind (e.g., by observing the drone’s motion as

measured by its IMU or optical ﬂow sensor). In this work,

we seek to leverage sensors that directly measure wind with

adequate accuracy and frequency to improve multirotor drone

control.

Conventional airﬂow sensors. One of the most com-

monly used sensors for measuring airspeed on ﬁxed-wing

platforms is the pitot-static tube [14]. However, they suffer

from high lag and typically have settling times on the

order of seconds [15], making them unreliable for unsteady

ﬂow measurements. Moreover, conventional one-dimensional

pitot-static tubes do not provide estimates of ﬂow direction,

and are in fact insensitive to changes in direction of up to

15◦-20◦[16]. This renders them unsuitable as an omnidirec-

tional ﬂow sensor. Another conventional airﬂow sensor is

the hot-wire anemometer [17]. While these sensors have a

signiﬁcantly faster response (on the order of milliseconds

or even microseconds [18]), their form factor is not well-

suited for multirotor applications. Speciﬁcally, they typically

require large, heavy, and expensive circuitry, and are also

fragile due to the unshielded microscale wire [19].

Airﬂow sensors for multirotor drones. Motivated by

the challenges with conventional airﬂow sensors, there have

been efforts to develop custom airﬂow sensing technology

for multirotor drones. For example, ﬂow speed (but not

direction) can be estimated by measuring the deﬂection of a

single sensing element [20], [21]. The whisker-like sensors

used in [22] can resolve airﬂow direction from deﬂection

using multiple sensing elements; however, [22] does not

provide a characterization of the temporal characteristics (lag

and frequency) of the sensors. Differential pressure probes

have also been used for resolving the wind vector [23], [24].

However, these have relatively high errors in angle estimation

(12◦error in [24]), and are based on principles used by

conventional airﬂow sensors (which suffer from high latency

and low temporal resolution). Another sensing modality

uses fast-response multi-hole pressure probes (MHPPs) [25],

which yield wind magnitude and direction estimates of 1 m/s

and 5◦respectively. However, these sensors are unable to

measure outside a 90◦cone of acceptance, and also utilize a

large probe whose length is that of the drone itself [26]. In

this work, we utilize recently developed MEMS-based hot-

wire sensors, which afford a small form factor (millimeter

scale), fast response (>500Hz), and accurate wind speed

and direction estimation.

II. SYSTEM OVERVIEW

In this section, we provide an overview of the FlowDrone

architecture and its main components; this architecture is

illustrated in Fig. 2.

Drone hardware. The FlowDrone platform is built on top

of the Holybro X500 Kit with a Pixhawk 4 autopilot. The

drone is equipped with an accelerometer and magnetometer,

as well as an onboard Holybro M8N GPS for compass

readings (the primary yaw reference), and a PX4FLOW

Smart Camera and LIDAR-Lite v3 for optical ﬂow and height

measurements. The GPS is not used as the ﬂights were

conducted indoors. These sensors are used by the Pixhawk

to calculate state estimates, which are then passed to an

onboard companion computer (a Raspberry Pi 4 8 GB Model

B) over the PX4-ROS2 bridge at 100 Hz. We highlight that

all sensing and computation is onboard, thus allowing for

future outdoor deployment.

Sensors and wind estimation. The key feature of the

FlowDrone is the use of MEMS hot-wire technology for

文档加载中……请稍候！
如果长时间未打开，您也可以点击刷新试试。

下载文档到电脑，查找使用更方便

10 玖币 0人已下载

立即下载

摘要：

FlowDrone:WindEstimationandGustRejectiononUAVsUsingFast-ResponseHot-WireFlowSensorsNathanielSimon,AllenZ.Ren,AlexanderPiqu´e,DavidSnyder,DaphneBarretto,MarcusHultmark,andAnirudhaMajumdarAbstractUnmannedaerialvehicles(UAVs)arendinguseinapplicationsthatplaceincreasingemphasisonrobustnesstoexternaldi...

展开>> 收起<<

FlowDrone Wind Estimation and Gust Rejection on UA Vs Using Fast-Response Hot-Wire Flow Sensors Nathaniel Simon Allen Z. Ren Alexander Piqu e David Snyder Daphne Barretto.pdf

共7页,预览2页

还剩页未读，继续阅读

声明：本站为文档C2C交易模式，即用户上传的文档直接被用户下载，本站只是中间服务平台，本站所有文档下载所得的收益归上传人(含作者)所有。玖贝云文库仅提供信息存储空间，仅对用户上传内容的表现方式做保护处理，对上载内容本身不做任何修改或编辑。若文档所含内容侵犯了您的版权或隐私，请立即通知玖贝云文库，我们立即给予删除！

FlowDrone Wind Estimation and Gust Rejection on UA Vs Using Fast-Response Hot-Wire Flow Sensors Nathaniel Simon Allen Z. Ren Alexander Piqu e David Snyder Daphne Barretto

相关推荐

开通VIP享超值会员特权

作者详情

相关内容

热门标签

举报选择: