Fields2Cover An open-source coverage path planning library for unmanned agricultural vehicles Gonzalo Mier1João Valente2Sytze de Bruin3

2025-05-06 0 0 4.14MB 8 页 10玖币

侵权投诉

Fields2Cover: An open-source coverage path planning library for

unmanned agricultural vehicles

Gonzalo Mier1João Valente2Sytze de Bruin3

Abstract— This paper describes Fields2Cover1, a novel open

source library for coverage path planning (CPP) for agricul-

tural vehicles. While there are several CPP solutions nowadays,

there have been limited efforts to unify them into an open

source library and provide benchmarking tools to compare their

performance. Fields2Cover provides a framework for planning

coverage paths, developing novel techniques, and benchmarking

state-of-the-art algorithms. The library features a modular and

extensible architecture that supports various vehicles and can

be used for a variety of applications, including farms. Its

core modules are: a headland generator, a swath generator,

a route planner and a path planner. An interface to the Robot

Operating System (ROS) is also supplied as an add-on. In

this paper, the functionalities of the library for planning a

coverage path in agriculture are demonstrated using 8 state-

of-the-art methods and 7 objective functions in simulation and

ﬁeld experiments.

Keywords – Agricultural Automation, Software Architecture

for Robotic and Automation, Field Robots

I. INTRODUCTION

In developed countries, there is a shortage of skilled

workers to operate agricultural machinery [1]. This shortage

can be alleviated with the development of autonomous ma-

chinery. Unlike manually operated machinery, autonomous

vehicle operations need meticulous planning beforehand. The

problem of determining a path to cover a ﬁeld is known as

coverage path planning (CPP). CPP is of high importance

for cleaning [2], surveillance robots [3], lawn mowers [4],

and agricultural vehicles [5], where it has been addressed in

several works.

Whilst there have been many efforts, most of the (par-

tial) CPP solutions have not been released as open-source

software thus hindering more rapid advances in CPP by the

scientiﬁc community. The packages shown in Table Iare

the only open-source software to the best of our knowledge.

Note that the software packages listed in Table Isolve the

CPP problem partially, but require several modiﬁcations in

order to be customized to different unmanned vehicles and

applications.

This paper aims to ﬁll the above mentioned gap by

proposing and releasing to the community an open-source

CPP library for ﬁeld coverage. The library was designed

1Gonzalo Mier with the Laboratory of Geo-Information Science and

Remote Sensing, Wageningen University & Research, Wageningen, The

Netherlands gonzalo.miermunoz@wur.nl

2João Valente with the Information Technology Group, Wageningen

University & Research, Wageningen, The Netherlands

3Sytze de Bruin with the Laboratory of Geo-Information Science and

Remote Sensing, Wageningen University & Research, Wageningen, The

Netherlands

1https://github.com/Fields2Cover/Fields2Cover

focusing in four modules that are the core of CPP solutions: a

headland generator, a swath generator, a route planner, and a

path planner. Each module includes at least one state-of-the-

art method and one objective function. The library currently

only supports convex ﬁelds on arable farmland. Regardless,

there is an urgent need for an open source software solution

to ﬁll the existing gap in the CPP problem in agriculture.

The ultimate goal of the library is to ease the state of-the-

art algorithm benchmark and to accelerate CPP research and

application.

A. Related work

Owing to the non-holonomous nature of agricultural ve-

hicles, a region of the ﬁeld known as headlands must be

reserved for turning the vehicle. The most basic approach

is to allocate a constant width area around the ﬁeld. This

strategy allocates a large amount of space to a poor yield

area. Depending on how the swaths are arranged, some

headland areas are parallel to the swaths and hence they

are not needed for turning. By only constructing headlands

along the ﬁeld edges where turns are made, the area reserved

for them can be minimized [5], [17].

Swaths are generated in the inner ﬁeld, which is the

remaining region after subtracting the headlands. In two-

dimensional planar ﬁelds, a reference line can be applied as

a guide for the generation of swaths, where each parallel

creates a swath [5], [17], [18]. This line can be chosen

for convenience or by an algorithm such as brute force

or a meta-heuristic. Oksanen [5] describes a driving angle

search strategy that requires fewer iterations than brute force

search but it does not guarantee ﬁnding the global minimum.

Objective functions such as the number of turns or the sum

of swath lengths are used to determine optimality in swath

generation [17].

The distance [17] and time [19] required to cover the

ﬁeld are affected by the order of the swaths. A route is the

sequence of the swaths to cover. The Boustrophedon order,

which travels the swaths sequentially from one side of the

ﬁeld to the other, and the snake order, which skips one swath

at each turn and returns through the uncovered swaths, are

popular preset routing patterns [20]. Objective functions such

as distance, number of rotations, or time necessary to traverse

the ﬁeld [17], [19] are minimized by ﬁnding the optimal route

through meta-heuristics [21].

A path is composed of the swaths of a route connected

by turns, forming a continuous line along which the vehicle

will drive. Dubins’ [22] or Reeds-Shepp’s [23] curves are

turns that minimize the path length of the turns. These

arXiv:2210.07838v2 [cs.RO] 17 Feb 2023

TABLE I: Comparison between coverage path planning open-source software solutions. Repositories (rows) are compared in

terms of (1) available documentation (Docs); (2) Computation of exact solutions rather than using a discretizing grids (No

grid); (3) Support for non-holonomous vehicles in turns; (4) The option to reserve maneuvering space at the ﬁeld boundaries

(Headlands support); (5) The possibility to modify the objective function; (6) Applicability for agricultural ground robots.

Package name Docs No grid

used

Non-

holonomous

Headlands

support

Customizable

objective

function

Terrestrial

agricultural

vehicles

RJJxp/CoveragePlanning [6] 747 7 7 7

Nobleo/full_coverage_path_planner [7] 4747 7 7

Ipa320/ipa_coverage_planning [2], [8], [9] 4 4 4 7 7 7

Ethz-asl/polygon_coverage_planning [10], [11] 4 4 7 7 7 7

Irvingvasquez/ocpp [12], [13] 747 7 7 7

Greenzie/boustrophedon_planner [14] 74 4 4 7 7

Ipiano/coverage-planning [15], [16] 4 4 7 7 47

Fields2Cover 4 4 4 4 4 4

curves are made by either curve segments or straight lines.

The main problem is that there is an instantaneous change

of curvature at the transition point between two segments.

Techniques such as numerical integrators [24] or clothoids

[25] are employed to smooth the turn to avoid the curvature

discontinuity. Furthermore, to navigate from a swath to

the headlands, turns such as non-uniform rational B-spline

(NURBS) curves [26] can be adopted.

CPP problems are composed of numerous sub-problems,

several of which have received special attention in literature.

For example, Spekken [21] presents an approach for calcu-

lating the coverage path in undulating terrain that however

does not consider turns between rows or headland creation.

Nilsson [27] and Nørremark [28] divide the CPP problem

into two major modules: Field Partitioning/Representation,

where the distribution of headlands and swaths in the ﬁeld

is set up, and Route Planning, which determines the opti-

mal order of travelling the swaths within sub-ﬁelds. In the

latter framework, each module has more than one function,

increasing the complexity of comparing multiple variations

of the module.

B. Existing open-source software

There have been web applications, such as GAOS [18],

that allowed farmers to design or adapt coverage paths with

a user-friendly interface. Many of such web applications,

despite being a great help to the farming community, have

been developed in collaboration with companies, restricting

the possibility to release the code to the public domain.

The currently existing open source CPP repositories are

listed in Table I. Although seven other projects were found,

none of them can be adopted for farming purposes with

ground robots. As mentioned above, ground robots in agri-

culture are generally non-holonomous, so turning maneuvers

must be planned to move from one swath to another. Unfor-

tunately, some packages [6], [10], [12], [15] only compute

the route to cover a region. These packages are designed

for quadrotors [10], [12] or for indoor robots [6]. However,

the code needs to be modiﬁed to support path generation

for non-holonomous robots. A special case of CPP is the

Nobleo package [7] which, although the vehicle used is non-

holonomous, uses a grid to deﬁne the nodes that should

be covered at least once. In agriculture it is important to

reduce the damage caused by the wheels of the vehicle,

so it is not recommended to cover the same swath several

times [7] or to cross through the main ﬁeld [6], [8], [10].

On the other hand, Greenzie [14], which was developed for

lawn mowers, is the only package that supports headlands,

along with Fields2Cover. Unlike arable farming, mowers are

constrained to avoid repeated tracks for ﬁeld trafﬁc, thus

the coverage path is created with random sweep angles. For

this reason, Greenzie does neither provide an optimizer nor

an objective function for planning the swaths. In contrast,

Ipiano [15] provides an interface to change the objective

function used by its optimizer, but here no headland support

is offered. Fields2Cover is the only software solution that

provides algorithms to create a coverage path for terrestrial

agricultural robots, including optimizers and objective func-

tions to generate the best path, headland support and turn

planning.

C. Contributions

The main contributions of this paper and the Fields2Cover

library are:

1) A publicly-available library (Fields2cover) providing

connectable modules to address CPP problems with

unmanned agricultural vehicles. Those modules can be

effortlessly customized for other CPP problems.

2) Benchmark tools for quantitative comparison between

the CPP algorithms and approaches.

3) A quantitative comparison using 38 convex ﬁelds be-

tween eight state-of-the-art CPP approaches/methods

and seven objective functions.

4) Experiments with a commercial unmanned agricul-

tural vehicle demonstrating Fields2Cover’s capability

to provide real-world solutions.

5) Contribute to building a research community by pro-

viding an accessible platform for discussing novel

ideas, solving issues and sharing new variants of the

problem.

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：

Fields2Cover:Anopen-sourcecoveragepathplanninglibraryforunmannedagriculturalvehiclesGonzaloMier1JoãoValente2SytzedeBruin3AbstractThispaperdescribesFields2Cover1,anovelopensourcelibraryforcoveragepathplanning(CPP)foragricul-turalvehicles.WhilethereareseveralCPPsolutionsnowadays,therehavebeenlimitede...

展开>> 收起<<

Fields2Cover An open-source coverage path planning library for unmanned agricultural vehicles Gonzalo Mier1João Valente2Sytze de Bruin3.pdf

共8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！

Fields2Cover An open-source coverage path planning library for unmanned agricultural vehicles Gonzalo Mier1João Valente2Sytze de Bruin3

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: