HYBRID INDOOR LOCALIZATION VIA REINFORCEMENT LEARNING-BASED INFORMATION FUSION Mohammad Salimibeniy Arash Mohammadiy

2025-05-08 0 0 2.47MB 5 页 10玖币

侵权投诉

HYBRID INDOOR LOCALIZATION VIA REINFORCEMENT LEARNING-BASED

INFORMATION FUSION

Mohammad Salimibeni†, Arash Mohammadi†

†Concordia Institute of Information Systems Engineering (CIISE), Montreal, Canada

ABSTRACT

The paper is motivated by the importance of the Smart Cities (SC)

concept for future management of global urbanization. Among all

Internet of Things (IoT)-based communication technologies, Blue-

tooth Low Energy (BLE) plays a vital role in city-wide decision

making and services. Extreme ﬂuctuations of the Received Signal

Strength Indicator (RSSI), however, prevent this technology from

being a reliable solution with acceptable accuracy in the dynamic in-

door tracking/localization approaches for ever-changing SC environ-

ments. The latest version of the BLE v.5.1introduced a better possi-

bility for tracking users by utilizing the direction ﬁnding approaches

based on the Angle of Arrival (AoA), which is more reliable. There

are still some fundamental issues remaining to be addressed. Exist-

ing works mainly focus on implementing stand-alone models over-

looking potentials fusion strategies. The paper addresses this gap

and proposes a novel Reinforcement Learning (RL)-based informa-

tion fusion framework (RL-IFF) by coupling AoA with RSSI-based

particle ﬁltering and Inertial Measurement Unit (IMU)-based Pedes-

trian Dead Reckoning (PDR) frameworks. The proposed RL-IFF

solution is evaluated through a comprehensive set of experiments il-

lustrating superior performance compared to its counterparts.

Index Terms—Angle of Arrival (AoA), Bluetooth Low Energy

(BLE), Convolutional Neural Network (CNN), Indoor Localization,

Internet of Things (IoT).

1. INTRODUCTION

It is expected that in near future Smart Cities (SCs) [1] will ﬂour-

ish across the globe aiming at efﬁcient management and control of

global urbanization to assist in having more reliable and safe soci-

eties. Internet of Things (IoT) [1], plays a vital role in developing

different SC services by providing an interactive infrastructure. The

large volume of data provided by IoT devices can assist in providing

services to control and solve the different SC problems, especially

in indoor environments. The key underlying challenge here is to

efﬁciently fuse and process such large amounts of IoT data. Rein-

forcement Learning (RL) techniques [2] are attractive solutions for

this challenge (to fuse IoT data) as RL approaches can learn and

adapt to the environmental changes and can directly learn from his-

torical SC data where having a uniﬁed model is too expensive to de-

velop. Among different SC services, Contact Tracing (CT) [3, 4] for

pandemic control has become of paramount importance especially

given recent COVID-19 pandemic. The main building block of au-

tonomous CT frameworks is the proximity/localization module. Ex-

isting CT models are, primarily built for outdoors and can not offer

scalable solutions for indoor environments. On the other hand, exist-

ing indoor localization techniques, typically, utilize a single technol-

ogy or are developed based on a single processing model. Therefore,

such solutions commonly suffer from different issues such as sensi-

tivity to multi-path effects, noise, ﬂuctuations of received signal, and

frequency/phase shifts. Capitalizing on issues associated with stand-

alone solutions and the importance of the indoor localization topic,

especially during this pandemic area, the paper focuses on multi-

modal and Reinforcement Learning (RL)-based information fusion

techniques that allow for simultaneous integration of different tech-

nologies and processing solutions.

Literature Review: To achieve a reliable localization service, one

major approach is to leverage the information from multi-sensor sys-

tems to improve the system robustness, accuracy of the prediction

and enhance the detection range. Information fusion is one major

approach in multi-sensors IoT networks which can be categorized

into traditional methods and Machine Learning (ML)-based [5, 6]

approaches. Probabilistic fusion, evidential belief reasoning fusion,

fuzzy theory, and tensor fusion are among the main traditional infor-

mation fusion strategies [7]. In [5], practical use of ML methods as

an information fusion technique is reviewed and different informa-

tion fusion levels including signal-level, feature-level and decision-

level are discussed. Data labeling and training to fuse data obtained

from multiple data sources, however, is a time consuming and costly

procedure [8, 9]. RL-based methods [10–13] can address such is-

sues for resource-constrained IoT edge devices. Generally speaking,

RL targets providing human-level adaptive behavior by construction

of an optimal control policy [14]. The main underlying objective

is learning (via trial and error) from previous interactions of an au-

tonomous agent and its surrounding environment [2, 15–19]. For

example in [10], to address the problems related to the weighted fu-

sion method with ﬁxed weight allocation in decision level fusion,

a deep RL multi-modal decision-making fusion weight allocation is

developed. The powerful decision-making ability of deep RL is used

in this approach to mitigate the problems related to the traditional

fusion models. Liu et al. [20] proposed a deep RL multi-type data

fusion framework to solve the issues related to the complicated stock

market environment and fusing different data types for algorithmic

trading services. In this approach, an static Markov Decision Pro-

cess (MDP) is deﬁned for the proposed signal fusion strategy and

RL is mainly used when the main fusion task is done on temporal

features of stock data and others to make trading decisions. The idea

mainly is not following a RL-based fusion strategy and classical fu-

sion solutions are applied in this work. Another priori knowledge

RL-based information fusion method for multi-sensors for air com-

bat data is proposed in [13]. In this paper, RL is used to ﬁnd the

coefﬁcients for the fusion of the information for different sensory

data input. An static MDP is also considered to be used in this ap-

proach to be able to apply RL on the fusion phase. Whereas some

of the above-mentioned works may lack timeliness and are not ap-

plied for localization purposes, indoor localization tasks deal with

a time-varying system. There are different sensor fusion localiza-

tion approaches [21–23] proposed in literature, yet the fusion sec-

tion is not implemented based on RL approaches, mostly relying

on the traditional sensor fusion techniques or supervised ML tech-

arXiv:2210.15132v1 [cs.AI] 27 Oct 2022

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：

HYBRIDINDOORLOCALIZATIONVIAREINFORCEMENTLEARNING-BASEDINFORMATIONFUSIONMohammadSalimibeniy,ArashMohammadiyyConcordiaInstituteofInformationSystemsEngineering(CIISE),Montreal,CanadaABSTRACTThepaperismotivatedbytheimportanceoftheSmartCities(SC)conceptforfuturemanagementofglobalurbanization.AmongallInte...

展开>> 收起<<

HYBRID INDOOR LOCALIZATION VIA REINFORCEMENT LEARNING-BASED INFORMATION FUSION Mohammad Salimibeniy Arash Mohammadiy.pdf

共5页,预览1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！

HYBRID INDOOR LOCALIZATION VIA REINFORCEMENT LEARNING-BASED INFORMATION FUSION Mohammad Salimibeniy Arash Mohammadiy

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: