IEEE Consumer Electronics Magazine - July/August 2022 - 66

Industrial Automation
accuracy. To overcome the loss in accuracy, we design and integrate a KF that learns the
dynamics of the robot motion to provide highly accurate tracking. Our system achieves
centimeter-level accuracy for the two aisles with RMSE of 0.35mand 0.37 m, respectively.
Further, compared with ML only localization systems, we achieve a significant reduction
in RMSE by 28.5% and 54.3% within the two aisles.
& IN THE RECENT decade, advances in industrial
automation have gained a lot of attention from
academia and industry. Leading to the evolution
of the fourth industrial revolution, known as
Industry 4.0.1,2,3 The advances are made in the
computation and the communication aspects of
industrial automation targeting robotics and
material handling. Realizing Industry 4.0 requires
the design of autonomous robots working in
coordination to perform various tasks. One of
the preliminary requirements for the robot is to
localize itself within the environment with very
high accuracy for coordination and carrying out
tasks. Localization requires information from
various kinds of sensors. Sensors such as LiDAR,
vision, and wireless are most commonly used for
localization.4,5,6,7 Kukkala et al.8 detailed discussion
regarding advanced driver assistance systems
utilizing wireless, radar, and LiDAR-based
hardware and software components for self-driving
vehicles. They present the use of different
consumer sensors for tasks like localization,
mapping, and navigation. Lemley et al.9 used
deep learning for designing smarter and intelligent
consumer devices and software services.
In recent years, with the maturity of consumer-grade
sensor technology, localization
algorithms are heavily studied. However, reliable
localization is still challenging for accurate and
reliable autonomous design. The challenges
arise due to the unpredictable environmental
conditions for both indoor and outdoor applications.
Traditional approaches to wireless localization
use triangulation and trilateration to
estimate the position of the client. Using such an
approach for localization requires accurate
knowledge of the wireless channel model is
required. Millimeter-wave channel in an indoor
environment suffers from multipath propagation.10
To model the wireless channel, detailed
knowledge of the electromagnetic characteristics
from all scatters is required. Such modeling
of indoor wave propagation at 60 GHz is nontrivial
and involves custom hardware design. For
this, the mmWave equipments used in our
approach are the consumer-grade routers capable
of communicating at 60 GHz frequency. Such
consumer hardware is designed to communicate
and provides coarse-grained channel state information
due to cost-effective antenna array
design. We propose a machine learning and Kalman
filter-based indoor localization system to
alleviate the challenges associated with channel
modeling and irregular signal strength. The system
learns the mapping between the complex
wireless radio features and the distances to provide
dynamic position tracking.
Many data augmentation techniques are
heavily employed for machine learning applications
like image classification, object detection,
and object recognition, where the images in the
training datasets are rotated, and cropped. The
augmentation is performed for two primary reasons:
first, to increase the size of the training
dataset without physically collecting more data,
which saves the data collection time, and second,
to increase the robustness of the ML model
for inference on unseen test data and has shown
drastic and significant improvement in the learning
capabilities of neural networks. Most of the
data augmentation techniques are for image/
vision-based datasets.11 In our design, the features
are the wireless SNR information and not
image pixels and cannot be used on wireless features.
To improve ML-based localization models,
we propose a new data augmentation technique
for wireless features. To perform data augmentation,
we synthetically generate another set of
radio map, assuming a large-scale shadow fading
in the 60-GHz propagation model.10
Table 1 summarizes the localization performance
using different sensor modalities, namely,
global positioning system (GPS), LiDAR, and
vision. Simultaneous localization and mapping
66
IEEE Consumer Electronics Magazine
IEEE Consumer Electronics Magazine - July/August 2022

Table of Contents for the Digital Edition of IEEE Consumer Electronics Magazine - July/August 2022
