Architecture and Challenges of IoT in developing an

Infrastructure for Robot Taxi

A. Vijay
Assistant Professor (Sr. Gr) Ayesha Mustafa Wardah Afzal
Sri ramakarishna engineering college Computer Science Electrical Engineering
Coimbatore, Tamil Nadu, India [email protected] [email protected]
[email protected]

Maria Tariq Kousla K Surname

Aatika Shehzad Electrical Engineering Professor at Depaartment of computer
Doctor of Vertinary And Medicine [email protected] science and engineering organization
[email protected] Kongu Engineering College
Perundurai, India
[email protected]

Abstract—Autonomous robot taxis are the future self-driven The first level starts from Level 0 which is fully manual and
robot taxis that can be operated on their own without the help the last level is Level 5 which is fully autonomous [2]. And
of any driver. They are superfast in speed and would not these levels have been defined by the Society of Automotive
cause an accident. In short, they are ideal supercars. The Engineers (SAE).
motive behind their development is to provide such vehicles
An autonomous robot taxi is designed in order to move
that could easily bring a revolution in the traffic system of
on your provided instructions [3]. Let’s say you wanna go to
the world. To achieve that much better performance,
a specific destination. You just have to give the car,
industrialists have started doing their work in this field by
instructions and then it will follow your orders.
solving the issues that they are facing in the current cars to
obtain their desired results. Full cooperation of such vehicle Self-driven is the synonym of the word autonomous.
with other vehicles, road and the whole network is needed in Both of them can be used for this kind of robot taxis [4].
successful launch of such vehicles. One of the major Usually, for self-driven vehicles, there remains the need for a
challenges in it is to convert the already formulated cars into person who resides in the car just to take the charge while the
such type of robot taxis. Then, for the smooth working of car moves on its own. They are level 3 cars but autonomous
these robot taxis, the obstacles around it must be evaluated. cars fall under the level 5 category [5].
Many other challenges along with solutions are presented in
this paper. A brief analysis about the sensors used in the
development of autonomous robot taxis are mentioned in the Now, come to the working of autonomous robot taxis.
architecture section. Moreover, this article also deals with the First of all, it relies on sensors as well as processors. Then
architecture of such robot taxis. It is also predicted that these come the algorithms which are also necessary to provide
vehicles will be widely used all over the world in the near them with a pattern to follow. And machine learning also
future. plays a significant role in it [6].

Keywords—IoT (Internet of Things), GPS (Global

Positioning system), Robot taxi, OP code (Operational Code)

I. INTRODUCTION

An autonomous robot taxi is a type of self-driven car that

checks its surroundings and operates without any assistance
from a human being [1]. And, the passenger has nothing to
do in order to operate it. It will move on its own. Just like a
normal car, it can be used to go anywhere you want. But
without any driver.
Basically, there are six levels of driving automation that
has been adopted by the US Department of Transportation.
 Chassis:
II. ARCHITECTURE This is the final stage of the process. At this stage, some
machinal components are involved. Basically, they are
mounted over chassis. They involve motors like a gear
motor, pedal motor, etc. And the control module will manage
them [12].
Now, let’s discuss the functionality of some important
sensors of the autonomous robot taxis.
Ultrasonic sensors:
Some waves within the range of 20 to 40k Hz are
required to operate these sensors [13]. Actually, such kinds
of waves originate from a membrane known as a magneto-
resistive membrane. The basic purpose of that membrane is
to check the distance toward a certain object. This is done by
checking the actual time of flight at which the wave emerges
to the signal of the echo [14]. Their range is limited and that
is almost 3 meters. While the result is updated every 20
microseconds. All of these sensors have the ability to check
First of all, it is important to know that we can change the direction at a very low range. That is why many other
any kind of taxi into an autonomous robot taxi. It can be done sensors are required. In case you want to check the whole
by adding sensors and some other components in it. And by view. But many sensors if work together will cause different
doing so, we are giving it the ability to make decisions errors regarding the range. So, the optimal solution for it is to
regarding its surroundings. The following figure illustrates get a code that can assist in discarding other echoes. In an
the changes which can be done in order to make it an Autonomous robot taxis, such sensors will assist in checking
autonomous robot taxi. the nearby distance in a limited range. Another benefit of
these sensors is that they work under every condition whether
Perception: it be a dusty storm or a rainy night [15].
At this stage, sensors are used to sense the environment III. CHALLANGES
around autonomous vehicles. These sensors are also used by
autonomous vehicles to detect their own position. These A. Data storage and processing:
sensors include Radar, Lidar, GPS, etc. The main goal of According to IDC (International Data Corporation), the
these sensors is to evaluate the information and after that, this global sphere of data will reach 163ZB (Trillion Gigabytes)
info is passed to the next module [7]. The next module is the by 2025. Autonomous vehicles generate 5TB to 20TB of data
recognition module. This module includes a control system for each vehicle per day [16]. This includes cameras, which
and an unknown obstacle recognition (UOR) framework as typically generate 20 to 60Mbps depending on the quality of
well. Similar other modules are also available for gathering the images captured, which can range from standard
info. After seeking the required info, it is then fused and definition to higher definition, as well as sonar (10 to
circulated towards the nest stage which is the planning stage 100kbps), radar (10kbps), LIDAR (Light Detection And
[8]. Ranging) systems (10 to 70Mbps), and GPS (Global
Planning: Positioning System) (10 to 70Mbps) [17]. The key is to
ensure that sensors collect the correct data and that it is
At this stage, the data which is gathered at the previous immediately processed, securely stored, and transferred to
stage will be used. This stage will tell autonomous robot taxi other technologies in the chain. The training stage depends
about how to move. Just like the brain in the human body. on GPUs along with a robust solution of computing which
All major decisions are taken at this stage. Proper planning assists in providing low latency and throughput from the
about movement is involved at this stage [9]. It will let the medium of storage for training. It appears to be extremely
robot taxi decide which path should be followed and which difficult to collect valuable data from devices such as
hindrance should be avoided through the info collected from cameras, radar, GPS, etc. in order to train neural networks
the previous stage. Details about the traffic in the [18]. These facts prove that data storage and processing is a
surroundings will also be useful for this stage. Real-time map big challenge regarding autonomous vehicles based on IoT.
data is also needed at this stage. An important module which
is called the data module will be used to save the info about B. Detecting malware:
the previous mistakes. Then, this info will be used in the near As the number of autonomous vehicles is increasing day
future[10]. by day, the number of cybercrimes also started increasing.
Hackers try to exploit all autonomous vehicles. They use
malware to harm them [19]. And by doing so, they are
Control: causing hindrance in the performance of these vehicles. X86
processors are used in some autonomous vehicles and it has
At this stage, data taken from the planning module is become the most vulnerable to cyber attackers to attack it.
utilized [11]. Some important functions regarding the actual And due to such kinds of attacks, their performance has been
control of autonomous robot taxi are performed at this stage. halted. As we know, there are some restrictions related to
These functions include applying brakes and accelerating etc. software and hardware in X86 processors, so the malware in
them can be detected through the .exe file. By running
the .exe file, we can do an analysis of this system to check IV. SOLUTIONS
whether malware is present or not. Its major purpose is to
check the Opcode. Another way to check the presence of A. Data storage and processing:
malware is to check through the entropy method. The following are some machine learning techniques that
Manufacturers use various methods but deep learning cannot can be employed in IoT-based autonomous vehicles.
be used to detect the presence of malware [20].
• For feature extraction, use SIFT (scale-invariant feature
C. Emergency Breaking: transform):
Here are some details about an incident that occurred in SIFT algorithms recognize and analyze items in photos.
the United States on August 14, 2018 at 10 p.m. When the The three points of a triangular symbol, for example, are
woman stepped into the street in Tempe, Arizona, it was entered as features. Using those locations, an automobile can
completely dark. According to the police, a car driving by at quickly detect the sign.
the time – a self-driven vehicle from US ride-hailing service
which is Uber -did not try to stop. That vehicle collide which • AdaBoost is used for data classification program
resulted in the first death of a person as well as the collapse This algorithm collects and categorizes data in order to
of the self-driving car. Meanwhile, sensors stopped working, improve vehicle learning and performance. It improves
making it impossible for them to detect a human being on a decision-making by combining numerous low-performing
cycle in the center of a road. According to recent studies of classifiers into a single high-performing classifier.
Uber, when the car is driven by a computer, emergency
braking is disabled in order to avoid unpredictable • Object recognition with TextonBoost
maneuvering that may confuse other traffic participants. The TextonBoost algorithm works in a similar way to
Instead, the safety driver must act on their own, relying AdaBoost, but it takes into account data from shape, context,
solely on their vision, because the system will not warn them. and appearance to improve texton learning (micro-structures
But here's something encouraging to back up the claim. in images). It combines visual data with common data.
Drivers in Germany are generally positive about self-driving
cars. According to a recent TUV-Rheinland study on the • Histogram of Oriented Gradients (HOG)
safety of self-driving cars, three-quarters of German drivers HOG makes it easier to analyze an object's location,
are willing to travel in one. Nonetheless, they are skeptical of known as a cell, to see how it changes or moves.
the implications for data protection, cybercrime, and road
safety in the face of increasing automation. This is due to a • YOLO (You Only Look Once)
number of outstanding technical challenges as well as a lack
Objects such as persons, trees, and automobiles are
of a legal framework [21].
detected and grouped using this technique. It assigns unique
D. Large Scale Integration: characteristics to each class of object it groups in order for
In IoT, scalability is defined as the capacity to accept the the automobile to recognize them. Identifying and classifying
change in order to fit in already defined architecture. This objects is easiest with YOLO [25].
means that you have to adapt to the changes accordingly Race to 5G:
[22]. It is crucial consideration of the design to assist too
many devices that vary from one another in terms of memory Because it does not require a carrier, 5G increases the
as well as processing. By doing so, the performance of the data possibilities for autonomous vehicles tenfold: most data
system is increased irrespective of the size. And the input and transmission can be done without wide area
performance can be improved. All of this can be done by network coverage [26]. We're talking about vehicle-to-
edge computing. This sort of computing aids in improving vehicle and vehicle-to-infrastructure connections at a speed
the services of networking as well as computing too [23]. and capacity that older generation networks couldn't match.

E. Management of Infrastructure: B. Detecting malware:

Sharing of the vehicle will be among the services offered Malware can be detected in the following ways:
by the IoT-based autonomous vehicles. These services Fusion Feature Eliminator: It pulls features of the
depend on a variety of applications, from those gathering program from three different layers of autonomous robot
fewer rates’ data rates and the non-real-time communications taxis. These three different layers include PE header, API,
to those requiring high-speed real-time communications. As and OpCode. In the first layer which is the PE header, crucial
a result, two techniques would be one of the IoT-based data is stored to identify malware; just like, the malware
autonomous vehicle’s major research challenges. Among import table has fewer symbols than conventional software
these two techniques, one is designing efficiently and the because most malware uses the runtime link mechanism.
other is intelligent routing management techniques. There is a difference between malware and regular software
Architecture standardization improves interoperability across which can be checked by accessing specific APIs. Because
multiple connected devices. An ideal IoT system includes they will analyze to check the behavior as well as features
management functions that are simple and scalable, too. One of the most widely used static properties for
integration of heterogeneous systems and logical effectively expressing program behavior patterns is the
applications, cross-domain interactions, and user-friendly opcode. As a result, using OpCode n-grams as a criterion will
applications and data analytic tools [24]. reveal crucial differences in malware and benign software
activities. The program fingerprint's basic representation is
the structure information, and the top layer abstraction is the
function information, both of which officially represent the
program. Behavior is the only thing that describes the
program from the inside. There are three different levels of find nearby vehicles or taxis and keep track of transits'
info that must be included while capturing the fingerprint in projected arrival times. Communication can be made with the
IOT-based autonomous vehicles. And that info is about: pedestrian through a pedestrian walking system and change
structure, function, and behavior. The important features of traffic lights to allow pedestrians to cross a road.
the header of PE define a program's structure, whereas
• Vehicle to network (V2N): The driver is informed about
crucial features of API indicate the reason for the program,
the condition of the weather by the weather prediction
and features from the OpCode describe the program's
department. And the road accident can also be informed in a
behaviors. Just to accurately identify the malware. The actual
similar way. Furthermore, a car can be linked to a
functionality of OpCode can be checked by doing the
smartphone. Then, the music of the vehicle can easily be
conversion of the EXE file to an ASM file. And after it that
controlled by the voice of the driver.
file can be extracted.
XGBoost-based Malware Detector: It includes a data E. Management of Infrastructure:
structure, distributed computing, parallel computing, algo A popular organization decides to set a standard for
optimization. As a result, many Kaggle tournament winners checking the infrastructure of autonomous vehicles. The
prefer to utilize the XGBoost detector due to its performance. name of that organization is NDS Association. NDS stands
Vlad Sandulescu, for example, won the KDD Cup 2016 for Navigation Data Standards. This organization has joined
event with the help of XGBoost. The goal of XGBoost is to all the location providers such as HD maps, navigation apps,
separate features on a regular basis in order to grow trees. etc. in order to make HD maps more reliable for autonomous
The sample's expected value is the sum of the scores of the vehicles. They are also working for providing a platform on
relevant leaf nodes in each tree. XGBoost is used in our which all the HD maps can easily be stored and updated
malware detection for self-driven vehicles that have an X86- accordingly. Accessibility can be increased if such a platform
based processor based on IoT devices [27]. will be formulated. The same data can be accessed from
anywhere. And it can be updated in the same way. This
C. Emergency Breaking: organization has defined the complete pattern for the storage
To counter this, we can use a system known as advanced of data. And that data storage pattern is to store the data in
driver assistance systems (ADAS) can be utilized. form of blocks and then attach a unique ID or number to each
Autonomous Emergency Braking is one of the most block. And the data can be accessed through some protocols.
important features of the ADAS system (AEB). AEB's main Same in the case of maps. Maps are accessed through the
job is to turn near-misses into close calls. AEB plays a key defined set of rules according to the location. The major
role in averting collisions and alerting the driver in dangerous benefit of the NDS map is that it will remove the time
situations. To put it another way, it detects danger and constraint. And it eliminates the need for a separate system of
applies the brakes for the driver. navigation. It is that much accurate that anyone can use it
easily. Issue of data translation and difference in formats has
Blindspot detection is a crucial technology for giving
been also overcome. Moreover, by using NDS maps,
360-degree coverage of the vehicle irrespective of its speed.
vehicles do not need to verify data separately because there is
This Volvo-developed device monitors traffic behind the
one trusted and centralized pool of data that is too reliable.
vehicle as well as what's approaching from the side.
Due to this, security has been enhanced and safety can also
Blindspot detection is also known as blindspot be ensured. All in all, it has provided a wonderful experience.
monitoring. It is categorized into two major categories. One [30]
is active blindspot monitoring and the other is passive Conclusion:
blindspot monitoring [28].
By analyzing the above details, we conclude that there is
Electronic Stability Control (ESC) is a crucial feature that
a lot of scope for autonomous taxis not only in the
aids in retaining the control for crucial driving circumstances
department of transportation but also according to the
by using automatic computer-controlled braking of
research perspective too. And the only useful means to take
individual wheels. The function is activated when the driver
benefit from it is to implement it in a secure way. This can
loses control of the vehicle [29].
only be done through certain protocols. Obviously, setting
D. Large Scale Integration: these protocols is not an overnight task. First of all, these
• V2V (vehicle to vehicle): A connection of type V2V protocols must be set at the global level and then awareness
allows to sharing of the required data among the vehicles. must be created among the masses related to these protocols.
This data includes sharing the dynamics among the vehicles. And above all, simpler negligence will create havoc.
These dynamics include sharing the speed and location. This Moreover, many problems can be solved by the usage of
kind of integration is too much important for vehicles that are these robot taxis. One of the biggest issues is traffic flow. It
needed in case of emergency. can only be solved through them. Parking issues can also be
solved through them. All in all, it will benefit us by making
• V2I (vehicle to infrastructure): A V2I connection assists traveling easier for everyone. Most importantly, they will
in the communication of vehicles with the whole help a lot in case of emergency. Clearing the mines will
infrastructure. Roads. And in return, it will assist in become easier with their usage. But apart from all these
comparing the traffic flow. Then, the vehicle will move benefits, there are some challenges in implementing this
accordingly. Sometimes, it also assists in checking where is technology like cyberattacks, etc. Maybe starting cost will
the petrol pump just to refill the petrol in the vehicle. remain an issue but not for a long time.
• Vehicle to pedestrians (V2P): By using the app on the References:
smartphone, a person can easily connect to the CV2X
network using a smartphone application. This app is used to
[1] Kehinde Lawal, Hamed Nabizadeh Rafsanjani, 2022, ‘Trends, [14] J. Levinson et al., ‘‘Towards fully autonomous driving: Systems and
benefits, risks, and challenges of IoT implementation in residential algorithms,’’ in Proc. IEEE Intell. Vehicles Symp. (IV), Jun. 2011,
and commercial buildings’, Volume 3, Issue 3, Pages 251-266, pp. 163–168.
[2] H. Sun et al., “Cloudeyes: Cloud-Based Malware Detection with [15] C. Berger and B. Rumpe. (2014). ‘‘Autonomous driving-5 years after
Reversible Sketch for Resource-Constrained Internet of Things (IoT) the urban challenge: The anticipatory vehicle as a cyber-physical
Devices,” Software: Practice and Experience, vol. 47, no. 3, 2017, pp. system.’’ [Online]. Available: https://arxiv.org/abs/1409.0413
421–41. [16] J. Kim, R. R. Rajkumar, and M. Jochim, ‘‘Towards dependable
[3] Weina Niu, Xiaosong Zhang, Xiaojiang Du, Teng Hu, Xin Xie, and autonomous driving vehicles: A system-level approach,’’ ACM
Nadra Guizani, “Detecting malware on X86 based devices” 2019, pp. SIGBED Rev., vol. 10, no. 1, pp. 29–32, 2013.
86-89 [17] J. Wei, J. M. Snider, J. Kim, J. M. Dolan, R. Rajkumar, and B.
[4] M. Shyam, R. SathyaVignesh, A. Sivakumar, J. Yogapriya, Litkouhi, ‘‘Towards a viable autonomous driving research platform,’’
“Vehicular Speed Regulation through Autonomous Braking System in Proc. IEEE Intell. Vehicles Symp. (IV), Jun. 2013, pp. 763–770.
Embedded with Smart Insurance Facility using Iot”, Volume-8, Issue- [18] M. R. Endsley, ‘‘Autonomous driving systems: A preliminary
5, 2020 naturalistic study of the tesla model s,’’ J. Cognit. Eng. Decision
[5] C.Alaoui, Z.M.Salameh,“On-Board Diagonostic and Rejenuvation Making, vol. 11, no. 3, pp. 225–238, 2017.
System for Electric Vehicles”, 58th Vehicular Technology [19] P. Lin, W. Ren, and H. Gao, ‘‘Distributed velocity-constrained
Conference VTC-2003-Fall (IEEE Cat No. 03CH37484)Vol.5 ,IEEE consensus of discrete-time multi-agent systems with nonconvex
Conference, 2003 constraints, switching topologies, and delays,’’ IEEE Trans. Autom.
[6] P. Sethi, S.R. Sarangi, “Internet of things: architectures, protocols, Control, vol. 62, no. 11, pp. 5788–5794, Nov. 2017.
and applications”, J. Electr. Computer. Eng., 2017 (2017), Article [20] J. Qin, C. Yu, and H. Gao, ‘‘Collective behavior for group of generic
e9324035, 10.1155/2017/9324035 linear agents interacting under arbitrary network topology,’’ IEEE
[7] Yun Jin,Shengguan Li, Juan Li, HongbingSun,Yuanwang Wu, Trans. Control Netw. Syst., vol. 2, no. 3, pp. 288–297, Sep. 2015.
“Design of an Intelligent Active Obstacle Avoidance Car Based on [21] Gomes, L., “Hidden obstacles for Google’s self-driving cars,” MIT
Rotating Ultrasonic Sensors”, 8th Annual International Conference on Technology Review, August 28, 2014.
CYBER Technology in Automation, Control, and Intelligent Systems,
[22] Hammett, “Design by extrapolation: an evaluation of fault-tolerant
IEEE Conference, 2018.
avionics,” 20th Conference on Digital Avionics Systems, IEEE, 2001.
[8] K. M. Alam, M. Saini, and A. El Saddik, “Toward social internet of
[23] Honeywell, Malfunction Management Activity Area Report for AHS
vehicles: Concept, architecture, and applications,” IEEE Access, vol.
Health Management, AHS Precursor Task E report, DoT FHA
3, pp. 343–357, 2015.
Publication FHWA-RD-95-047, November 1995.
[9] G. Mao, Z. Zhang, and B. D. O. Anderson, “Cooperative content
[24] Hume, D. (1910) [1748]. An Enquiry concerning [22] Human
dissemination and offloading in heterogeneous mobile networks,”
Understanding. P.F. Collier & Son. ISBN 0-19- 825060-6.
IEEE Transactions on Vehicular Technology, vol. 65, no. 8, pp.
6573–6587, Aug 2016. [25] ISO 26262, Road vehicles – Functional Safety, International Standard,
First Edition, Nov 15, 2011, ISO
[10] W. Yao, Z. Deng, and L. Zhou, ‘‘Road curb detection using 3D lidar
and integral laser points for intelligent vehicles,’’ in Proc. Joint 6th [26] Kalra, Nidhi and Susan M. Paddock. “Driving to Safety: How Many
Int. Conf. Soft Comput. Intell. Syst. (SCIS), 13th Int. Symp. Adv. Miles of Driving Would It Take to Demonstrate Autonomous Vehicle
Intell. Syst. (ISIS), Nov. 2012, pp. 100–105. Reliability?” Santa Monica, CA: RAND Corporation, 2016.
[11] D. Liu, X. An, Z. Sun, and H. He, ‘‘Active safety in autonomous land [27] Kane, Chowdhury, Datta & Koopman, "A Case Study on Runtime
vehicle,’’ in Proc. Workshop Power Electron. Intell. Transp. Syst. Monitoring of an Autonomous Research Vehicle (ARV) System," RV
(PEITS), Aug. 2008, pp. 476–480. 2015.
[12] X. Li, Z. Sun, D. Cao, D. Liu, and H. He, ‘‘Development of a new [28] Kianfar, R., Falcone, P. & Fredriksson, J., “Safety verification of
integrated local trajectory planning and tracking control framework automated driving systems,” IEEE Intelligent Transportation Systems
for autonomous ground vehicles,’’ Mech. Syst. Signal Process., vol. Magazine, Winter 2013, pp. 73-86.
87, pp. 118–137, Mar. 2017. [29] Koopman & Wagner, “Challenges in Autonomous Vehicle Testing
[13] J. Ziegler et al., ‘‘Making Bertha drive—An autonomous journey on a and Validation,” SAE World Congress, April 2016.
historic route,’’ IEEE Intell. Transp. Syst. Mag., vol. 6, no. 2, pp. 8– [30] Learner, P., “The hurdles facing autonomous vehicles,” Automobile,
20, Apr. 2014. Jun. 22, 2015.