DRACo-SLAM Distributed Robust Acoustic Communication-efﬁcient SLAM for Imaging Sonar Equipped Underwater Robot Teams John McConnell Yewei Huang Paul Szenher Ivana Collado-Gonzalez and Brendan Englot

2025-04-27 0 0 1.45MB 8 页 10玖币

侵权投诉

DRACo-SLAM: Distributed Robust Acoustic Communication-efﬁcient

SLAM for Imaging Sonar Equipped Underwater Robot Teams

John McConnell, Yewei Huang, Paul Szenher, Ivana Collado-Gonzalez and Brendan Englot

Abstract— An essential task for a multi-robot system is gener-

ating a common understanding of the environment and relative

poses between robots. Cooperative tasks can be executed only

when a vehicle has knowledge of its own state and the states of

the team members. However, this has primarily been achieved

with direct rendezvous between underwater robots, via inter-

robot ranging. We propose a novel distributed multi-robot

simultaneous localization and mapping (SLAM) framework for

underwater robots using imaging sonar-based perception. By

passing only scene descriptors between robots, we do not need

to pass raw sensor data unless there is a likelihood of inter-robot

loop closure. We utilize pairwise consistent measurement set

maximization (PCM), making our system robust to erroneous

loop closures. The functionality of our system is demonstrated

using two real-world datasets, one with three robots and

another with two robots. We show that our system effectively

estimates the trajectories of the multi-robot system and keeps

the bandwidth requirements of inter-robot communication low.

To our knowledge, this paper describes the ﬁrst instance of

multi-robot SLAM using real imaging sonar data (which we

implement ofﬂine, using simulated communication). Code link:

https://github.com/jake3991/DRACo-SLAM.

I. INTRODUCTION

Underwater robotics has proliferated over the past decade,

supporting a variety of tasks, including ship hull inspec-

tion, harbor security, maintenance-inspection-repair (MIR),

and intelligence surveillance reconnaissance (ISR). How-

ever, autonomous underwater vehicles (AUVs) are limited

in their sensing capabilities. Firstly, the global positioning

system (GPS) cannot be used due to the attenuation of GPS

signals in the water. While AUVs can surface to collect

GPS measurements, this is inefﬁcient and impossible during

under-ice missions or certain tactical situations. Further, if a

small, lightweight, low-power perceptual package is desired,

underwater lidar may be infeasible, and the use of sonar

may be required, especially in turbid or dark environments.

While the capabilities of sonar-based systems have expanded

to include fully autonomous operation, the primary area of

contribution is still focused on single agents executing their

missions. However, real-world problems often beneﬁt from

a team of robots, a multi-robot system. When considering

tasks ranging from autonomous exploration to ISR, a multi-

robot system can reduce time, energy expenditure or create

redundancy when operating in an adversarial environment.

A fundamental capability for both single and multi-robot

systems is state estimation. When operating in unknown en-

J. McConnell, Y. Huang, P. Szenher, I. Collado-Gonzalez and B. Englot

are with Stevens Institute of Technology, Hoboken, NJ, 07030, USA

{jmcconn1,yhuang85,pszenher,icollado,benglot}@stevens.edu

This research was supported in part by ONR Grant N00014-21-1-2161

and NSF Grants IIS-1652064 and IIS-1723996.

(a) Our experimental

“AUV”: a custom-

instrumented BlueROV2.

(b) Satellite image of the SUNY Mar-

itime marina.

from the perspective of one robot. This robot’s poses are shown as

black dots connected by several factors: sequential scan matching factors

(SSM) in green, non-sequential scan matching factors (NSSM, intra-

robot loop closures) shown in red, inter-robot (IR) loop closures in blue

and partner robot (PR) factors in purple. This robot’s map is shown in

red with blue points merged from the other robot in the mission.

Fig. 1: Distributed multi-robot SLAM overview: (a) the robot

used in our experiments, (b) one of the settings for our experiments,

and (c) a representative experimental result are shown.

vironments, simultaneous localization and mapping (SLAM)

is utilized to estimate vehicle location and provide situational

awareness. For underwater robots, SLAM has typically been

implemented to localize a single vehicle. Recent works on

SLAM for underwater multi-robot systems consider both

direct encounters (e.g., inter-robot range measurements) and

indirect encounters (e.g., commonly observed targets in the

survey area) as a means for fusing robot state estimates

[1], [2]. There are some fundamental challenges to imple-

menting such systems. Firstly, acoustic communication links

have notably low bandwidth, limiting information exchange

between robots. Another issue is the nature of sonar as a

arXiv:2210.00867v1 [cs.RO] 3 Oct 2022

tool for perception. Sonar has a low signal-to-noise ratio,

low resolution, and often lacks 3D information. Moreover,

there are no (or poor) initial conditions relating the robots

in a multi-robot system without GPS.

In this work, we propose a system for distributed multi-

robot SLAM that uses an imaging sonar’s perceptual data

as a basis for state estimation, and we validate it over real-

world datasets. In addition to utilizing real perceptual data,

we do not provide any initial conditions relating the robots a

priori, and we do not rely upon inter-robot ranging, permit-

ting an exclusive reliance on indirect encounters between

robots. Lastly, we consider the bandwidth limitations of

wireless acoustic communications and design our system for

compatibility with realistic constraints (we note that beyond

bandwidth, there are additional limitations associated with

real-world acoustic comms [3]). Our key contributions are:

•The ﬁrst underwater multi-robot SLAM system devel-

oped for use with real imaging sonar data.

•A high-performance pipeline for robustly identifying,

registering, and rejecting outliers of indirect encounters

using real sonar data with no initial conditions.

•A strategy to manage bandwidth usage, making real

world application with acoustic modems a possibility.

•A validation of our system on two real-world data-sets.

In the following sections, we discuss related work, mathe-

matically deﬁne our problem, present our system and results.

II. RELATED WORK

A. Underwater Multi-Robot SLAM

In the underwater space, inter-robot constraints generally

come in two forms [2]: direct encounters where robots

observe one another via acoustic ranging, and indirect en-

counters, where robots observe the same targets in the envi-

ronment and may derive inter-robot measurement constraints

relating one another. We note that direct encounters require

synchronized clocks, which may not be practical over long

periods without GPS clock corrections due to clock drift.

We ﬁrst consider [4], where a mobile base station is used

to localize a team of low-cost vehicles lacking perceptual

sensors using an acoustic beacon. A similar concept is

considered in [5], except instead of a team of robots, a

leader-follower arrangement is used. [6] considers inter-

robot ranging without ﬁxed acoustic beacons and performs

a simulation study comparing the use of ﬁxed beacons to a

cooperative localization solution. [7] considers an algorithm

for processing inter-robot acoustic pulses in a distributed

manner. [1] proposes a pose-graph-based method for cooper-

ative localization of a team of robots using dead-reckoning,

GPS, and inter-robot ranging. The outcome of this work is a

system where a robot maintains an understanding of its state,

and the team’s state. Moreover, when GPS measurements

occur at the surface, their effect is shared across robots. [2]

integrates perception into the above, and commonly observed

features are shared and integrated into robot pose-graphs.

[2] assumes that robots share point-landmark observations

in their survey area, but the framework is only tested in

simulation. Further, due to GPS, each robot is effectively

localized in a common frame, enabling the sharing of range-

bearing measurements to commonly observed targets without

a need to solve the complex data association problem. Recall,

GPS may not always be available due to under-ice operations

or tactical situations (e.g., jamming). [8] considers the robot

map merging problem using only similarity in feature space,

in this case, ship hull curvature. While this work is able to

merge trajectories and lower the data transmission require-

ments between robots, it requires a highly descriptive feature

vector. It is only validated over a single dimension, whereas

we consider a 3DOF system.

There are some notable examples of underwater multi-

robot SLAM using cameras [9]–[12] and cameras with other

navigation sensors [13]. Cameras, however are not robust

in all water conditions. While these works are informative,

because we operate in turbid conditions with low visibility,

we do not consider them further.

Another area to consider is the concept of multi-session

SLAM, where a robot is provided with a prior SLAM run

and merges the map it builds with the prior run as the mission

progresses. This is examined for ship hull inspection [14],

[15], bathymetric mapping [16], ship-wreck reconstruction

[17] and environmental monitoring [18]. However, multi-

session does not account for communicating information

between robots, as it is a single agent system with prior

information. Additionally, it may be the case that the robot

has some knowledge of its location in the prior map, con-

straining the inter-robot loop closure search space.

In contrast to the above, we utilize no notion of an initial

guess relating robot reference frames. Second, we consider

the bandwidth limitations of real acoustic modem hardware

and take steps to manage network utilization. Lastly, we

implement a fully functioning system to detect and estimate

indirect encounters in sonar data: inter-robot loop closures.

B. Place Recognition With Sonar

Place recognition (loop closure) is fundamental for a

perception-driven multi-robot system. Place recognition has

been widely studied in LIDAR sensing [19]–[21]. These

works assemble a 2D bird’s eye view image of a 3D LIDAR

scan with coarse discretization to compare scenes. They also

derive a 1D descriptor to support scene search and retrieval.

Place recognition has also been studied with sonar-

equipped AUVs. Firstly, [22] considers building scene graphs

to compare scenes and evaluate rigid-body transformations

but requires at least ﬁfteen objects in each scene to run.

Machine learning has also been used to support this task in

[23], [24]. However, few public sonar datasets exist, and this

is often a research area unto itself. Iterative closest point

(ICP) based loop closure is used in our prior work [25]

to support single-agent active SLAM. This work uses sonar

derived point clouds with ICP; when ICP provides a trans-

form between keyframes, we estimate overlap between the

point clouds. Point cloud overlap, then pairwise consistent

measurement set maximization (PCM) [26], are used to reject

outlier loop closures. Inliers are integrated into the graph-

based pose SLAM solution. We note the extension of the

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：

DRACo-SLAM:DistributedRobustAcousticCommunication-efcientSLAMforImagingSonarEquippedUnderwaterRobotTeamsJohnMcConnell,YeweiHuang,PaulSzenher,IvanaCollado-GonzalezandBrendanEnglotAbstractAnessentialtaskforamulti-robotsystemisgener-atingacommonunderstandingoftheenvironmentandrelativeposesbetweenrobo...

展开>> 收起<<

DRACo-SLAM Distributed Robust Acoustic Communication-efﬁcient SLAM for Imaging Sonar Equipped Underwater Robot Teams John McConnell Yewei Huang Paul Szenher Ivana Collado-Gonzalez and Brendan Englot.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！

DRACo-SLAM Distributed Robust Acoustic Communication-efﬁcient SLAM for Imaging Sonar Equipped Underwater Robot Teams John McConnell Yewei Huang Paul Szenher Ivana Collado-Gonzalez and Brendan Englot

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: