Wi-Closure Reliable and Efﬁcient Search of Inter-robot Loop Closures Using Wireless Sensing Weiying Wang1 Anne Kemmeren2 Daniel Son1 Javier Alonso-Mora2 Stephanie Gil1

2025-05-06 0 0 3.2MB 7 页 10玖币

侵权投诉

Wi-Closure: Reliable and Efﬁcient Search of

Inter-robot Loop Closures Using Wireless Sensing

Weiying Wang1, Anne Kemmeren2, Daniel Son1, Javier Alonso-Mora2, Stephanie Gil1

Abstract—In this paper we propose a novel algorithm, Wi-

Closure, to improve computational efﬁciency and robustness of

loop closure detection in multi-robot SLAM. Our approach

decreases the computational overhead of classical approaches

by pruning the search space of potential loop closures, prior to

evaluation by a typical multi-robot SLAM pipeline. Wi-Closure

achieves this by identifying candidates that are spatially close

to each other by using sensing over the wireless communica-

tion signal between robots, even when they are operating in

non-line-of-sight or in remote areas of the environment from

one another. We demonstrate the validity of our approach in

simulation and hardware experiments. Our results show that

using Wi-closure greatly reduces computation time, by 54% in

simulation and by 77% in hardware compared, with a multi-robot

SLAM baseline. Importantly, this is achieved without sacriﬁcing

accuracy. Using Wi-closure reduces absolute trajectory estimation

error by 99% in simulation and 89.2% in hardware experiments.

This improvement is due in part to Wi-Closure’s ability to

avoid catastrophic optimization failure that typically occurs with

classical approaches in challenging repetitive environments.

I. INTRODUCTION

Loop closure detection has been widely studied as a fun-

damental aspect of Simultaneous Localization and Mapping

(SLAM) [1], [2]. The location estimate of the robot drifts

over time due to the noise in the on-board odometer and loop

closure detection is essential to correct for this drift by recog-

nizing previously visited places. Without such corrections, the

world as perceived by the robot may diverge substantially from

reality. Similarly, if multiple robots intend to collaborate, they

require a shared situational awareness being consistent with re-

ality, as obtained by multi-robot SLAM. The key to obtaining

this shared understanding are inter-robot loop closures. Where

regular loop closures constrain the positions of one robot itself,

1John A. Paulson School of Engineering and Applied Sciences, Harvard

University, Allston, MA 02134, USA

2Faculty of Mechanical, Maritime and Materials Engineering, Technical

University of Delft, 2628 CD Delft, The Netherlands

the inter-robot loop closure deﬁnes spatial relations between

pairs of robots. These inter-robot loop closures enable robots

to merge local sensor data into a shared model of the world

and obtain relative locations.

A common method that robots use to ﬁnd inter-robot loop

closures is place recognition. However, place recognition re-

mains challenging in practice, especially when the environ-

ment has repetitive elements [3] and when communication

between robots is intermittent. We introduce Wi-Closure to

address two persistent problems in this setting. First, since

robots do not know each other’s location, they may mismatch

similar-looking scenes that they have encountered in different

locations – a problem also referred to as perceptual aliasing

[1]. Second, during the short intervals that communication

between robots is established, feeding a large set of inter-robot

loop closures into the multi-robot SLAM pipeline puts a large

strain on computational resources [4]. Repetitive elements

further increase computation by falsely recognizing more inter-

robot loop closures. Previous work introduced pairwise con-

sistency maximization (PCM) to prevent scene mismatching

by identifying false inter-robot loop closures [5]. However,

recent research demonstrates that if repetitive elements are

present, catastrophic failure of the SLAM algorithm can occur

even if using PCM [6]. A more robust solution to perceptual

aliasing is tracking all possible (mis)matches, resulting in

various hypotheses of what the world looks like [7]. Un-

fortunately, working with multiple hypotheses is costly since

multiple-hypothesis tracking and planning are computationally

complex [8]. This makes these methods less viable for real-

time execution on commonly available robot hardware.

Our approach Wi-Closure is a computationally lightweight

method that robustly ﬁnds inter-robot loop closures in per-

ceptually aliased environments. We use spatial information

from WiFi and ultra-wideband (UWB) communication sig-

Fig. 1: Wi-Closure efﬁciently ﬁnds locations where robots’ trajectories overlap, as indicated by the yellow area. Only inter-robot loop closures at these

locations need to be processed by the multi-robot SLAM pipeline. This increases robustness against perceptual aliasing and decreases overall computation of

the pipeline.

arXiv:2210.01320v2 [cs.RO] 23 Nov 2022

nals to identify where robots’ trajectories are close. WiFi

is an electromagnetic wave, and thus the receiving robot

can locally derive the direction or Angle of Arrival (AOA)

to the transmitting robot from the phase information [9].

Similarly, commercial UWB devices measure time-of-ﬂight to

estimate distance. Importantly, sensing through the commu-

nication signal has wide applicability in this setting since it

passes through obstacles and thus works in non line-of-sight

situations [10], and it doesn’t require the robots to identify

each other through vision-based methods, e.g. using Apriltags

[11]. Our method solely uses spatial information and thus it

can work seamlessly together with existing place recognition

methods based on appearance information. As depicted in

Fig. 1, the Wi-Closure algorithm is used at the start of the

multi-robot SLAM pipeline.

In order to achieve good performance, Wi-Closure must also

address a major challenge to sensing over the communication

signal; namely, it must address multipath propagation of the

wireless signal. Multipath refers to the phenomenon where the

signal bounces off of various objects to arrive at the receiver

from different angles. Consequently, the AOA measurement

may include multiple directions, of which at most one is the

direct-line path to the other robot. We address this issue with

PCM, since only the true direct paths will give consistent

pairs of AOA measurements over time. In our hardware

experiments, after collecting 4 AOA measurements with in

total 3 direct paths and 17 multipaths, we are able to accurately

distinguish all direct paths from the multipaths.

Our numerical and hardware experiment results demonstrate

that our method efﬁciently prunes the search space of loop

closure candidates by 99% in simulation and 78.7% in hard-

ware experiments, and increases robustness against perceptual

aliasing by rejecting up front inter-robot loop closures between

distinct places and reducing absolute trajectory estimation

error by 99% in simulation and 89.2% in hardware results.

We summarize the contributions of this paper as follows:

1) We introduce a resource efﬁcient approach, Wi-Closure,

to detect inter-robot loop closures in perceptually aliased

environments, based on spatial information from the

communication signal. It can work in tandem with

existing place recognition methods.

2) We address the challenging situation of multipath prop-

agation of the communication signal with PCM.

3) We demonstrate the merits of our approach in terms

of robustness against false inter-robot loop closures

and improved computation time in simulation with the

KITTI dataset and in hardware experiments.

II. RELATED WORK

For decades, the majority of research on loop closure

detection has focused on a single robot [12], [13]. Recently

however, loop closure detection algorithms are being adapted

to ﬂeets of robots, to ensure reliable and efﬁcient retrieval

of shared map and location estimates [5], [14]. We leverage

previous work on sensing over the communication signal to

simultaneously address two open problems: 1) computation to

match large trajectories is high, and 2) place recognition easily

mismatches trajectories in repetitive environments.

Wireless sensing Extensive research has shown that we

can obtain spatial information from wireless signals [9], [15].

Many works use UWB sensors to obtain ranging information

between two robots by measuring the time-of-ﬂight of the

ultra-wideband signal. [16], [17] use the ranging information

amongst robots to improve the joint position estimate even

without being in line of sight of each other. Recently, [10] also

introduced sensing direction from the WiFi communication

signal to the robotics community, requiring only a single

WiFi antenna and movement of the robot. These innovations

avoid the need of bulky equipment and anchors as used

in classical works to estimate position, which come with

additional infrastructure requirements [18].

Range-only SLAM Previously, [19] used UWB sensors in

a multi-robot SLAM setting coined range-only SLAM, where

distance measurements are directly used as inter-robot loop

closures. This avoids the problem of perceptual aliasing, but

it only introduces connections between the maps of the robots

where the robots are communicating. In realistic scenarios

the communication is intermittent, and trajectories can overlap

in places where communication is unavailable and where the

position estimate is uncertain due to odometer drift. Additional

place recognition increases the accuracy of the map by match-

ing these overlapping locations. To our knowledge, we are the

ﬁrst to speed up place recognition using ranging and direction

information from the communication signal.

Computation in loop closure Researchers sought to reduce

computation of loop closure detection, e.g. with easily ob-

tainable ORB features for vision-based approaches [20], and

efﬁcient look-up trees to match scenes [12]. Unfortunately,

these methods may result in mismatched maps in perceptually

aliased environments [6]. In [21] the authors consider sampling

a subset of most informative inter-robot loop closures to reduce

overall time consumption. However, the authors also note that

the performance guarantee of their sampling method decreases

if a scene can be potentially matched to many others - i.e. when

there is substantial perceptual aliasing.

Perceptual aliasing Although repetitive scenes are per-

vasive in many environments, classical place recognition

approaches ﬁnd it notoriously difﬁcult to deal with them.

Researchers have focused on simultaneously representing all

possible matches as multiple hypotheses in one framework

[22]. However, to properly use these multiple hypotheses to

determine the best course of action for the robot, we need

computationally expensive methods such as data-association

belief space planning (DA-BSP) [8], [23]. In DA-BSP the

computation time scales exponentially with the hypotheses.

We observe that many methods have a trade-off between

robustness against perceptual aliasing and computation: in-

creased robustness requires large computation, while compu-

tationally efﬁcient methods decrease robustness or perform

worse in repetitive environments. Our approach instead aims to

improve both computation and robustness against perceptually

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：

Wi-Closure:ReliableandEfcientSearchofInter-robotLoopClosuresUsingWirelessSensingWeiyingWang1,AnneKemmeren2,DanielSon1,JavierAlonso-Mora2,StephanieGil1AbstractInthispaperweproposeanovelalgorithm,Wi-Closure,toimprovecomputationalefciencyandrobustnessofloopclosuredetectioninmulti-robotSLAM.Ourapproa...

展开>> 收起<<

Wi-Closure Reliable and Efﬁcient Search of Inter-robot Loop Closures Using Wireless Sensing Weiying Wang1 Anne Kemmeren2 Daniel Son1 Javier Alonso-Mora2 Stephanie Gil1.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！

Wi-Closure Reliable and Efﬁcient Search of Inter-robot Loop Closures Using Wireless Sensing Weiying Wang1 Anne Kemmeren2 Daniel Son1 Javier Alonso-Mora2 Stephanie Gil1

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: