Edge Computing Implementation

Submitted by: Likhit Gatagat

2022MT13057
Bits, Pilani, WILP
Email:
BITS [email protected]
SSZG586 – Edge Computing Assignment
Pilani
May 3, 2024 1
Problem Statement
From the given examples choose any two areas and come up with solution to implement it.
Please include different servers and devices are getting involved.

Selected Area: Autonomous Vehicles

Edge computing plays a crucial role in autonomous vehicles by enabling real-time processing of
sensor data. Cameras, LiDAR, radar, and other sensors on the vehicle generate a vast amount of
data that needs to be processed quickly to make split-second decisions. Edge computing
platforms installed within the vehicle process this data locally to ensure timely responses,
improving safety and efficiency.

NOTE: This file contains solution for Autonomous Vehicles, solution for another chosen area
(Healthcare Monitoring) is in another file named-
“2022mt13057_EdgeComputing_Assignment1_Health_Monitoring.pdf”
1. Edge Computing in Autonomous Vehicles
Autonomous vehicles like self-driving cars generate massive amounts of data from onboard
sensors like cameras, LiDAR, radar, and ultrasonic sensors. This data needs to be processed in
real-time to allow the vehicle to understand its environment and make split-second driving
decisions. Sending all this data to the cloud for processing would introduce too much latency.

Edge computing addresses this challenge by processing data locally on devices within the
vehicle or on edge data centers close to the vehicle. This allows much faster data processing
with lower latency. For a self-driving car traveling at high speeds, reducing latency from over
100 milliseconds in the cloud to under 10 milliseconds on the edge can mean the difference
between a safe stop or a collision.

Edge computing enables self-driving cars to make time-sensitive driving decisions by running AI
algorithms like computer vision and deep learning models locally on edge hardware inside the
car. This real-time edge processing is critical for functions like obstacle detection, traffic sign
recognition, and planning safe routes and manoeuvres. By processing data at the edge,
autonomous vehicles can operate more safely and efficiently on the road.
Autonomous Vehicles Ecosystem

Edge computing plays a crucial role in

autonomous vehicles by enabling real-time
processing of sensor data. Cameras, LiDAR, radar,
and other sensors on the vehicle generate a vast
amount of data that needs to be processed
Question 2
quickly to make split-second decisions. Edge
computing platforms installed within the vehicle
process this data locally to ensure timely
responses, improving safety and efficiency.
Autonomous Vehicle

 Today, newly manufactured cars are fitted with

hundreds of sensors that constantly share data
related to location, entertainment, environmental
conditions, vehicle operation, and driving behavior as
frequently as several times per second, generating
immense volumes of data. According to a McKinsey
estimate, connected cars create up to 25 gigabytes of
data per hour.

 The enormous amount of data generated by vehicles

is supported by a combination of data processing and
controls, sensors, artificial intelligence, and 5G, all
enabled by edge computing infrastructures. The
continued evolution of these integrated technologies
is the catalyst for new business models and use cases.
It can also enhance road safety and create more
intelligent user-led driving experiences.
 Edge Computing Technologies in AV
 Sensor fusion and value aggregation: Vehicle sensors generate a large volume of data. While most data can be
processed on board, certain applications push content to the cloud. However, leveraging edge computing limits the
need to move data externally, reducing both transmission costs and minimizing sensitive data leaving the vehicle.

 Advanced Driver Assist Systems (ADAS) for increased vehicle safety: ADAS enables pedestrian detection, automatic
emergency braking, and blind spot detection. It uses radar/Lidar/camera sensors on powerful edge computing systems
to support real-time vision coprocessing and sensor fusion subsystems.

 V2X protocols for enhanced connected car ecosystem: Vehicle-to-everything (V2X) connection allows vehicles to speak
both to each and surrounding infrastructure. V2X relies on edge computing and 5G networks for data exchange
between vehicles and servers with protocols requiring access to edge-based IoT devices.

 Authentication for data privacy: Vehicle data breaches and transmission errors can compromise vehicle and occupant
safety; hence data privacy is critical. Safeguarding against hacking can be achieved with robust edge infrastructure and
high-speed, secure connectivity. Using edge servers facilitates real-time authentication, however specific
communication protocol must be established and authenticated between the host and server to maximize safety and
minimize latency.
 Edge Computing Technologies in AV
 Intuitive Infotainment systems for enhanced user experience: Customer purchasing decisions are greatly influenced by the
quality of the vehicle infotainment with both drivers and passengers seeking intuitive experiences. Edge computing is again
responsible for meeting such expectations by reducing latency to deliver real-time personalization, music, and video
streaming, plus seamless software updates.

 Adaptive & predictive vehicle maintenance: Edge infrastructure allows machine learning (ML) and artificial intelligence (AI)
to be used to develop predictive maintenance schedules as part of vehicle after-sales service. It further facilitates remote

 Smart Home and Vehicle Integration for self-parking: Integrating the smart home and car allows one to communicate with
the other through compute power at the edge. For example, those with home valet parking can leave their vehicle in the
driveway, allowing the car to park itself in the garage. Through this intelligence, cars can also synch with the owner’s
calendar and drive themselves out of the garage ready for the scheduled departure.

 Generic Rules and Machine Learning-based Monitoring and Alerting: Such intelligence allows fleet managers to set up
rules to efficiently run operations. A rule may categorize a car as “unavailable” around the time it is due for a service,
triggering an alert. The role of edge computing here is to process sensor data on board and assess pre-set rules using ML
algorithms, avoiding the need to access the cloud. The cloud, however, can be leveraged to train and sharpen the ML
algorithms or initially define rules to then push to the edge when updates are made.
Networking Requirements

Autonomous vehicles generate and consume huge amounts of data from sensors, cameras,
radars, lidars, and machine learning models. This requires high bandwidth connections to
transmit all that data efficiently.

However, bandwidth alone is not enough. The vehicles also need very low latency connections,
meaning the delay between data transmission and processing/action is minimized. With
autonomous driving where split-second reactions are necessary, latency becomes a major
factor.

For example, if an object suddenly appears in the vehicle's path, the data from the sensors
needs to be processed instantly by the AI system to decide the appropriate driving response.
Even a delay of a few milliseconds could make the difference between a safe stop or an
accident.
Networking Requirements

This is why 5G networks are enabling wider adoption of

edge computing for autonomous vehicles. 5G delivers
the high bandwidth needed for all the data, along with
the ultra-low latency required for real-time processing
and decision-making. The quicker data speeds and
responsiveness allow vehicles to offload compute-
intensive tasks to the edge of the network.

With 5G and edge computing combined, autonomous

vehicles can operate more safely, efficiently and reliably.
The high bandwidth means lots of data can be
transmitted, while the low latency means real-time
processing with near-instant response times. This
powerful one-two punch will accelerate the self-driving
car revolution.
Network Architecture

According to Navigant Research, 94.7 million vehicles

with self-driving capabilities will be sold annually
by 2035. This phenomenal growth of intelligent,
connected vehicles generates staggering amounts of
data that require robust infrastructure and a solid
safeguard framework, supported by a plethora of
technologies from big data and AI to cloud and edge
computing. This network architecture and its
various layers are what make it all possible.
Network Architecture

 The Cloud Layer is where enormous volumes of data are stored, aggregated, and processed. It is also
here that more complex computation takes place. The cloud is ideal for longer-term storage in which
latency is not critical to applications

 The Mobile Edge Computing (MEC) Layer sits between the cloud and vehicular layer ensuring data
exchange and extending cloud capabilities, moving them closer to the edge. Information on road
infrastructure, pedestrians, other vehicles, and the external environment are shared here together with
data caching and computation functions. The combination of MEC and AI/ML allows vehicles to access
real-time information on their surroundings, decreasing latency and reducing data traffic congestion
which is critical to safety.

 The Vehicular Layer refers to communications between the vehicle and objects, people, or
infrastructure in its immediate vicinity. Intelligent connected vehicles (ICVs) collect data from onboard
and surrounding sensors, cameras, radar, Lidar, GPS, etc sending data to the MEC layer to feed various
vehicle intelligence services including environmental and traffic conditions. Data exchange is made
possible via the 5G network and communication with onboard wireless devices. As an example, the
integrated vehicle radar can slow the vehicle automatically upon sensing traffic build-up.
Data Processing at the Edge

Autonomous vehicles rely on data processing at the edge to

operate safely and efficiently on the road. This involves
processing data from cameras, radar, lidar and other sensors in
real-time inside the vehicle to understand its surroundings and
make decisions.

One key aspect is image recognition using footage from cameras

mounted around the exterior of the vehicle. Advanced deep
learning algorithms can analyse these video feeds to detect and
classify objects like other cars, pedestrians, traffic signs, and road
markings. This enables the autonomous vehicle to understand
exactly what is around it at all times.

In addition to cameras, most self-driving cars integrate LIDAR and

radar to create a complete 3D view of the environment. LIDAR
uses laser beams to measure distance, while radar detects speed
and direction of movement. Sensor fusion combines data from all
these sources to achieve a comprehensive perspective.
Data Processing at the Edge

The vehicle then processes all this sensor data locally

using artificial intelligence to make critical real-time
decisions on braking, accelerating, steering, and
manoeuvring safely through traffic. Performing these
compute-intensive tasks on the edge reduces latency
versus transmitting data to the cloud. This ensures the
vehicle can react quickly enough to avoid collisions and
handle the dynamics of driving at high speeds.

Overall, localized data processing at the edge is essential

for autonomous vehicles to safely navigate roads with no
driver input. It enables split-second reflexes and
responses to handle the complexities of real-world
driving.
Machine Learning at the Edge

Autonomous vehicles rely heavily on machine learning

models to operate. These models are initially trained in
the cloud using large datasets. However, running
complex machine learning models within a vehicle in
real-time requires processing at the edge.

Edge computing allows machine learning models to be

run locally on devices within the vehicle. This reduces
latency and allows for near real-time processing. The
models can make decisions based on data from sensors,
cameras, and other inputs.
Machine Learning at the Edge

In addition to running pre-trained models, edge computing enables continuous learning.

As the vehicle operates, it collects huge amounts of new data from its sensors and
cameras. Edge computing allows this data to further train the models on-the-fly. The
models can continuously improve through incremental learning on new data.

Local edge computing power means the vehicle does not need to wait for round-
trip communication with the cloud to apply machine learning. This enables faster
response times for modelling of the vehicle's environment and decision-making
during autonomous driving. With edge computing, vehicles can leverage advanced
machine learning while operating safely and efficiently on the road.
Conclusion

Edge computing is rapidly emerging as a critical enabling technology for autonomous vehicles and
other advanced applications. As highlighted in this article, edge computing allows self-driving cars
to process huge volumes of sensor data and run AI algorithms in real-time. Without the ultra-low
latency and localized processing of edge computing, autonomous vehicles would not be able to
make quick driving decisions to handle complex road conditions safely.