Autonomous Vehicles

What is Artificial Intelligence?

John McCarthy, a computer scientist, coined the term ‘Artificial Intelligence’ in 1955. AI is
defined as the ability of a computer program or machine to think, learn and make decisions.
In general use, the term means a machine which mimics human cognition. With AI, we are
getting computer programs and machines to do what humans do. We are feeding these
programs and machines with a massive amount of data that is analysed and processed to
ultimately think logically and perform human actions. The process of automating repetitive
human tasks is just the tip of the AI iceberg, medical diagnostics equipment and autonomous
cars have implemented AI with the objective of saving human lives.

Artificial Intelligence and Autonomous Vehicles:

Why do you think AI has become so popular now, though it has been around since the 1950s?
Simply put, the reason behind this explosion of AI is the humongous amount of data that we
have available today. With the help of connected devices and services, we are able to collect
data in every industry, thus fuelling the AI revolution. While efforts are being made to
rapidly improve sensors and cameras to generate data with autonomous vehicles, Nvidia
unveiled its first AI computer in October 2017 to enable deep learning, computer vision and
parallel computing algorithms. AI has become an essential component of automated drive
technology and it is important to know how it works in autonomous and connected vehicles.

The Growth of AI in Automotive:

The automotive AI market reported that it is expected to be valued at $783 million in 2017
and expected to reach close to $11k million by 2025, at a CAGR of about 38.5%. IHS
Markit predicted that the installation rate of AI-based systems of new vehicles would rise by
109% in 2025, compared to the adoption rate of 8% in 2015. AI-based systems will become a
standard in new vehicles especially in these two categories:
1. Infotainment human-machine interface, including speech recognition and gesture
recognition, eye tracking and driver monitoring, virtual assistance and natural
language interfaces.
2. Advanced Driver Assistance Systems (ADAS) and autonomous vehicles, including
camera-based machine vision systems, radar-based detection units, driver condition
evaluation and sensor fusion engine control units (ECUs).
Deep learning technology, which is a technique for implementing machine learning (an
approach to achieve AI), is expected to be the largest and the fastest-growing technology in
the automotive AI market. It is currently being used in voice recognition, voice search,
recommendation engines, sentiment analysis, image recognition and motion detection in
autonomous vehicles.
How Does AI Work in Autonomous Vehicles?
Let us first look at the human perspective of driving a car with the use of sensory functions
such as vision and sound to watch the road and the other cars on the road. When we stop at a
red light or wait for a pedestrian to cross the road, we are using our memory to make these
quick decisions. The years of driving experience habituate us to look for the little things that
we encounter often on the roads — it could be a better route to the office or just a big bump
in the road.
We are building autonomous vehicles that drive themselves, but we want them to drive like
human drivers do. That means we need to provide these vehicles with the sensory functions,
cognitive functions (memory, logical thinking, decision-making and learning) and executive
capabilities that humans use to drive vehicles. The automotive industry is continuously
evolving to achieve exactly this over the last few years.
According to Gartner, by 2020, 250 million cars will be connected with each other and the
infrastructure around them through various V2X (vehicle-to-everything communication)
systems. As the amount of information being fed into IVI (in-vehicle infotainment) units or
telematics systems grows, vehicles will be able to capture and share not only internal system
status and location data but also the changes in its surroundings, all in real time. Autonomous
vehicles are being fitted with cameras, sensors and communication systems to enable the
vehicle to generate massive amounts of data which, when applied with AI, enables the
vehicle to see, hear, think and make decisions just like human drivers do.

AI Perception Action Cycle in Autonomous Vehicles:

A repetitive loop, called Perception Action Cycle, is created when the autonomous vehicle
generates data from its surrounding environment and feeds it into the intelligent agent, who in
turn makes decisions and enables the autonomous vehicle to perform specific actions in that
same environment. The figure below illustrates the data flow in autonomous vehicles:

Figure 1: AI Perception Action Cycle in Autonomous Cars

Let us break this process down into three main components:
Component 1: In-Vehicle Data Collection & Communication Systems
Autonomous vehicles are fitted with numerous sensors, radars and cameras to generate
massive amounts of environmental data. All of these form the Digital Sensorium, through
which the autonomous vehicle can see, hear and feel the road, road infrastructure, other
vehicles and every other object on/near the road, just like a human driver would pay attention
to the road while driving. This data is then processed with super-computers and data
communication systems are used to securely communicate valuable information (input) to the
autonomous driving cloud platform. The autonomous vehicle first communicates the driving
environment and/or the particular driving situation to the Autonomous Driving Platform.
Component 2: Autonomous Driving Platform (Cloud)
The Autonomous Driving Platform which is in the cloud contains an intelligent agent which
makes use of AI algorithms to make meaningful decisions. It acts as the control policy or the
brain of the autonomous vehicle. This intelligent agent is also connected to a database which
acts as a memory where past driving experiences are stored. This data along with the real-
time input coming in through the autonomous vehicle and the immediate environment around
it helps the intelligent agent make accurate driving decisions. The autonomous vehicle now
knows exactly what to do in this driving environment and/or particular driving situation.
Component 3: AI-Based Functions in Autonomous Vehicles
Based on the decisions made by the intelligent agent, the autonomous vehicle is able to detect
objects on the road, makeover through the traffic without human intervention and gets to the
destination safely. Autonomous vehicles are also being equipped with AI-based functional
systems such as voice and speech recognition, gesture controls, eye tracking and other
driving monitoring systems, virtual assistance, mapping and safety systems to name a few.
These functions are also carried out based on the decisions made by the intelligent agent in
the Autonomous Driving Platform. These systems have been created to give customers a
great user-experience and keep them safe on the roads. The driving experiences generated
from every ride is recorded and stored in the database to help the intelligent agent make much
more accurate decisions in the future.
This data loop, called Perception Action Cycle, takes place repetitively. The more the number
of Perception Action Cycles take place, that much more intelligent the intelligent agent
becomes, resulting in a higher accuracy of making decisions, especially in complex driving
situations. More the number of connected vehicles, more the number of driving experiences
are recorded, enabling the intelligent agent to make decisions based on data generated by
multiple autonomous vehicles. This means that not every autonomous vehicle has to go
through a complex driving situation before it can actually understand it.
Artificial intelligence, especially neural networks and deep learning, have become an
absolute necessity to make autonomous vehicles function properly and safely. AI is leading
the way for the launch of Level 5 autonomous vehicles, where there will be no need for a
steering wheel, accelerator or brakes.

 AI algorithms used in autonomous vehicles:

1. Route Planning and Control Algorithms:

 Traditional algorithms from computer science that are heuristic in nature can be used
for this task. These are algorithms like Bellman-Ford and Dijkstra’s algorithm
(Bugala, 2018). For these algorithms to work we need to have localization of the
vehicle during the whole time. Localization is accomplished through sensors such as
GPS as well as simultaneous localization and mapping (SLAM) techniques.

SLAM is used when there is no GPS availability such as underground or enclosed

spaces for example. SLAM generates a map of the environment and at the same time
estimates the state of a vehicle .SLAM is used in applications where the map is not
available and needs to be created. It uses sensors and special algorithms that create
models of the data in order to produce the map.

2.  Object Detection Algorithms:

Object detection is one of the most important tasks that AI has to handle in a moving
vehicle. These algorithms are an area of active research and they rely on different
sensors. Object detection can be based on cameras or lidars, radars and other type of
sensors. The algorithms used are normally deep learning algorithms which use some
type of a neural network to do the job.
One requirement for this task is that it needs to be fast.  The reason is because there is
a succession of images that need to be processed as the vehicle moves.

3. Decision Making Algorithms:

Decision making determines the actions of the vehicle based on information from
sensors. A vehicle constantly makes decision, based on its policy and the
environment. The algorithms used for decision making are the following:
● Decision Trees
● Support Vector Machine (SVM) Regression
● Deep Reinforcement Learning

Challenges of AI in Autonomous Vehicles:
Some of the challenges of using artificial intelligence algorithms for autonomous vehicles are
the same challenges that are universal for many other AI applications. The following
subsections describe these areas in detail.

● Real time response

Real time systems are specific class of embedded systems which have the
characteristics to produce outputs or reactions within certain time restrictions. They
have to be deterministic and minimalist in their design, so that they can always meet
the expected real time behaviour. For this to happen they often use special real time
operating systems (RTOS) or bare metal executives that interact directly with the
hardware, avoiding interpreted languages and dynamic memory allocation. In some
 cases, real time systems intentionally use a subset of a programming language in order
to guarantee speed and determinism.

● Computational complexity

Artificial intelligence algorithms, especially deep learning, require special hardware

solutions due to the amount of data and the complexity of computations. Hardware,
like graphic processing units (GPU) or tensor processing units (TPU) can be highly
optimized for fast parallel computation.  The speed comes at a price of higher energy
consumption and higher cost. Even with using specialized hardware there is still no
certainty that a particular algorithm will be able to reach a solution in real time
constraints.

● Black box behaviour

AI algorithms have been criticized for being harder to analyse than standard computer
algorithms. This comes from the fact that AI algorithms have a higher level of
complexity and rely on a lot of data. A complex neural network can perform several
AI tasks without understanding the process it controls.

● Accuracy and Reliability

The computer vision applications used in autonomous vehicles may not be ready for
prime time. The reason is that they can work in pristine conditions, but can fail with
some even small disturbances at the sensor inputs. The training of AI algorithms
happens slowly with training data that has certain characteristics.

Changing the data can make the classification and prediction algorithms change their
behaviour dramatically with catastrophic results. For example, a person who carries a
large bag, may not be perceived as a person by the vehicle.

● Safety

The complexity of AI can lead to problems with safety. A more complicated system is
one that is more difficult to develop, test and deploy.  This   is only part of the
problem. Another reality is that new safety standards are now emerging and not yet
adopted by the transportation industry

● Security and AI

AI systems are so central to the autonomous vehicle that their security is directly
related to robustness of the whole system. Machine learning systems are susceptible
to attacks and there have been many examples of such cases 
 A big threat for affecting how AI operates in an autonomous vehicle is through
perturbing sensor operation, which leads to changes in the sensor data stream and thus
can completely confuse the AI algorithms. 

● Ethics and AI

At this point, the ethical aspects of AI implementations in autonomous and semi-

autonomous vehicles are not mature and are not significantly developed. Ethical
values are a human quality, which is hard to formalize and implement in machines. In
other words, deciding what is right or wrong is kind of nebulous to a machine.