Software-Defined Vehicle

MAGAZINE Issue 1 | October / November 2024

Luminar looks
beyond hardware

Continental's cross-domain HPC slashes development cycles | SoundHound’s Chat AI reshapes UX |

Elektrobit talks open-source OS | SDVs pave the way for the AI-defined automaker |
The AI-defined automaker | Automated ADAS testing advances autonomy |
IN THIS ISSUE
05
Continental makes SDVs close to personal
computers on wheels

08

08 16 20
SoundHound: OEMs An open-source OS Software-defined
embrace AI-powered changes the value vehicle advances to
software-centric UX proposition of vehicles AI-defined automaker

12 16
SDV success
requires a
harmonised
approach

Copyright statement

© 2024 All content copyright Automotive World Ltd. All rights reserved.

This publication - in whole or in part - may not be shared, copied, reproduced, stored in a retrieval system,
or be transmitted in any form by any means electronic, mechanical, photocopying, recording or otherwise
without the prior permission of Automotive World Ltd.

October / November 2024 Software-Defined Vehicle Magazine 3

 24 28
May Mobility: AVs Luminar: software is
Automotive World must prioritise safety the key differentiator
1-3 Washington Buildings and accessibility in sensor tech
Stanwell Road, Penarth,
CF64 2AD, UK 20

www.automotiveworld.com
[email protected]

Registered number:
04242884
VAT number:
GB 815 220 173
24 32
DeepRoute.ai CEO:
CEO & Managing Director:
SDVs and AGI herald
Gareth Davies
‘the era of robots’

Contributors:
Will Girling
36
SDVs will be enabled
Megan Lampinen
40 by “deep integration”
Stewart Burnett
of generative AI
Pedro Moreno

Production:
Anmol Mothy
40
Automated ADAS
testing will take
© Automotive World Ltd 2024
autonomy into the
Cover image courtesy of Luminar future
4 Software-Defined Vehicle Magazine October / November 2024
Continental
makes SDVs
H
igh-performance
computers (HPCs) are a

close to central component

enabling the rise of
software-defined vehicles (SDVs).
This hardware provides the

personal processing and computing power

necessary for automakers to realise
innovative, software-driven features.
Combined with over-the-air updates,

computers HPCs can enable automotive to

iterate vehicle functionality
continuously in an equivalent
manner to the tech industry.

on wheels Today, automotive engineers

continue to chase higher system
performance while also seeking
reductions in overall cost and power
Continental’s work on high- consumption. Enhanced HPC
functionality could come increasingly
performance computer under the spotlight as an important
area for optimisation. As such, multi-
functionality and virtual testing national Tier 1 supplier Continental
could accelerate the development has been exploring not just how
hardware and software can be
cycle for SDVs. By Will Girling enhanced to unlock new functions
but also how to bring SDVs to the
market faster.

October / November 2024 Software-Defined Vehicle Magazine 5

 several functions across the same HPC,
Continental can deliver software
updates and new functions.

Unlocking new
business models
Although cross-domain HPCs hold
An example of an HPC—these units connect,
control, and enable a wide range of hardware
exciting potential, Schleicher
and software in SDVs emphasises that their practical
implementation is currently far from
simple. Certain functions could require
substantial graphics power or
Fragmentation to accompanying machine learning
capabilities on the device to run
consolidation properly. Discovering how disparate
domains running on the same HPC
Martin Schleicher, Head of Software interact will be an important learning
Strategy at Continental, explains to experience. “That’s one of the reasons
Automotive World that the use of HPCs in we’re doing this: to understand how
vehicles has historically been far we can take domain consolidation,”
fragmented. The first generations were he says.
kept separate from each other and
targeted specific domains, such as Schleicher anticipates that
cockpit and advanced driver assistance consolidating systems will not only
systems (ADAS). In May 2024, reduce vehicle weight and potentially
Continental became the first company shrink overall vehicle size over time
to buck this trend by implementing a but also unlock new business models.
cross-domain HPC in a real vehicle. The ultimate outcome would
effectively be general-purpose
The shift from domain to server-zone computers on wheels with sufficient
architectures enables vehicle compute power to add a wide variety
electronics to be grouped into sections of software-driven features over the
with their own ECUs and specific lifetime of a vehicle—a distinct
functions. For Continental, the critical monetisation opportunity that only
enabling component was the SDVs can unlock.
Snapdragon Ride Flex System-on-Chip,
which supports multi-modal workloads “By retaining the same hardware but
on a single hardware device. upgrading the functionality with
software, automakers can keep their
“Cross-domain HPCs mean we can run cars attractive and differentiate their
ADAS functions on a cockpit computer, brand,” he says. “The flexibility of a
for example, even though their cross-domain HPC will also allow them
functionalities are usually separate,” to move functions that don’t fit in one
says Schleicher. This includes complex ECU to another, meaning OEMs can
features like automatic parking using determine how and where their
ultrasonic sensors and surround view compute power is distributed.” Since
cameras. In addition to broadening everything in an electric SDV will be
system capabilities by consolidating drawing from the same power source—

6 Software-Defined Vehicle Magazine October / November 2024

 “ Continental wants to demonstrate
both what’s possible in terms of
hardware and what we can offer
automotive engineers as
standalone software

the battery—customisable efficiency will HPC: “We can accelerate the time-to-
also help maximise range, another market for new functions by allowing
important consumer consideration. engineers to test new software in a
virtual environment before it goes into
production.” Continental estimates that
Building enthusiasm for CAEdge can generate five million virtual
SDVs kilometres and 100PB of test data in just
one week. By facilitating parallel
By 2030, McKinsey & Co forecasts that hardware and software development,
global automotive software and the company claims that product
electronics will be worth US$462bn—up development cycles can be significantly
from around US$2.5bn in 2019. In the condensed. “A cockpit HPC, for
wake of the CrowdStrike system outage, example, might have taken 24 to 36
automotive players will be more aware months to validate. We can bring that
than ever that bugs and software errors down to 18 months,” states Schleicher.
can lead to disastrous results on a global
scale. With the consultancy also The road to SDVs will be challenging:
estimating that 90% of vehicles sold by automakers must balance functionality,
2030 will have connectivity features, upgradability, and compatibility while
some of which will be critical safety also considering how a system fits
functions, this is something the industry holistically within the dimensions and
cannot afford to risk. The challenge with specifications of a vehicle. Demand for
automotive becoming truly software- these time and cost saving tools already
defined is that development must exists, according to Schleicher, although
balance speed with safety: test cycles the industry might need to wait up to
must be as short as possible to be three years for a fully commercially
competitive without creating undue risk viable cross-domain HPC. “At this point,
because of unvalidated work. Continental wants to demonstrate both
what’s possible in terms of hardware
Through the Continental Automotive and what we can offer automotive
Edge Framework (CAEdge), a cloud- engineers as standalone software
based development environment, functions.” He hopes this technical
Schleicher states that automotive “showcase” will continue to build
engineers could examine, develop, and enthusiasm for the possibilities of SDV
maintain system software on a virtual technology in the coming years.

October / November 2024 Software-Defined Vehicle Magazine 7

SoundHound: OEMs
embrace AI-powered
software-centric UX
AI-powered digital assistants are emerging at the heart of
the software-defined vehicle. By Megan Lampinen

8 Software-Defined Vehicle Magazine October / November 2024

A
I-powered features promise
to reshape the in-vehicle
experience, and the rise of
the software-defined
vehicle (SDV) is making it easier than
ever for automakers to incorporate
them. The industry is increasingly
moving towards vehicles that use
software not only to manage their
operations but also add functionality
and enable new features. Generative
AI (GenAI) is one of the hottest
commodities on this front, SoundHound Chat AI for Automotive uses a proprietary approach to
particularly around voice interaction avoid misleading and unpredictable responses
with the vehicle’s digital assistant.
The promise on offer is one of smart,
natural interactions with an AI
assistant that not only hears but car’s mechanical operations at any
listens, understands, and responds. point, they can ask the assistant
about it, such as ‘Why do my brakes
feel sticky?’
Spotlight on
SoundHound “Our proprietary technology stands
out because it can manage the voice
California-based SoundHound’s Chat assistant’s responses across different
AI for Automotive was one of the first domains to consistently deliver
products to offer an in-vehicle voice accurate responses—whether for
assistant with GenAI capabilities. vehicle controls, real-time weather
“Chat AI integrates with third-party and travel updates, or complex GenAI
generative AI models, like ChatGPT, responses,” emphasises Zagorsek. To
enabling seamless conversations and select an appropriate response
quick, accurate responses—without quickly, the system draws on
any of those ‘I’m sorry I didn’t get proprietary SoundHound AI software
that’ moments,” says Chief Operating engineering technology called
Officer Michael Zagorsek. “You can CaiLAN (Conversational AI Language)
now speak naturally to the vehicle and the in-house machine learning
assistant and get immediate technology CaiNet (Conversational AI
responses from GenAI that work Network). CaiLAN builds knowledge
seamlessly alongside other real-time domains like weather, restaurants,
SoundHound domains.” traffic, and local search, while
CaiNET uses machine learning to
For example, a user may start by better understand queries and
asking what the weather is like in provide the right responses. CaiNET
Edinburgh. They can follow up with a can connect to SoundHound AI’s
query on the quickest route into the internal models as well as external
city, followed by a request for some models like ChatGPT. CaiLAN
picturesque stopping points along the controls and arbitrates the results to
way. If something feels ‘off ’ with the provide the best response to the user.

October / November 2024 Software-Defined Vehicle Magazine 9

 The range of potential use cases is responses to “a proprietary
vast. As well as trip planning, approach.” While still early days,
recommendations for food, and fuel momentum for its system is growing.
and parking suggestions, it could also
help with vehicle maintenance: Chat In February 2024, Chat AI entered
AI has a Vehicle Intelligence feature full production with Stellantis’ DS
that allows the user to access Automobiles, rolling out across 18
information directly from the car’s countries in 13 languages. Stellantis
manual just by speaking. In-car later added it to the Peugeot, Opel,
education and entertainment are and Vauxhall brands. Zagorsek
another potential area for growth, as suggests the recent product wins are
AI can create a dynamic in-car “testament to the success of the
experience for interactive games and technology and its popularity with
learning activities. For example, an drivers”. For certain previously
AI voice assistant could provide released Stellantis models,
occupants with a guided tour of what SoundHound Chat AI with
they are seeing out the window integrated ChatGPT will be available
based on topics of interest. on-demand and can be delivered
over-the-air (OTA) with no need to
visit a dealer.
An AI voice ecosystem
At the time of writing, SoundHound’
Hallucinations have been one of the is currently running pilots with
drawbacks with early applications of other unnamed OEMs in Europe and
AI-powered systems, but a growing anticipates a growing market in the
number of players are now claiming US market as well. In 2023, it
to have successfully tackled that commissioned a survey that found
problem. SoundHound is among 50% of regular drivers in the US are
them, attributing the elimination of likely or very likely to use an in-car
such misleading and unpredictable voice assistant with Gen AI

10 Software-Defined Vehicle Magazine October / November 2024

 ChatGPT launched in 2023 on all DS Automobiles
models with the DS Iris System

capabilities. “We can fairly assume Or even bigger. SoundHound also

that figure has risen by now due to provides voice AI solutions to
the increased popularity of AI restaurants and can be found in more
among consumers,” Zagorsek tells than 10,000 such locations, as well as
Automotive World. other customer-facing businesses. In
the past 12 months, it made two
As interest grows, so too will the significant acquisitions in the
competition. In Q1 2024, Cerence’s restaurant space with the purchase of
Chat Pro arrived on the first SYNC3 in December 2023 and Allset
Volkswagen Group models in Europe in June 2024.
and North America, offering very
similar capabilities and targeted at “Our vision is to build the voice
identical use cases. Deployments commerce ecosystem of the future,
across such big-name volume players whereby drivers and passengers will
as Stellantis and Volkswagen Group be able to use our in-vehicle voice
point to the mass market potential of assistant to complete all kinds of
AI voice technology, which is only at transactions,” says Zagorsek. This
the start of its journey. “The could include food ordering or
automotive industry is undergoing appointment scheduling, among
rapid change and beginning to other things, and would be entirely
embrace software-centric hands-free and directly from the
experiences fuelled by AI,” notes vehicle, television, or connected
Zagorsek. “We anticipate that many device. “This creates a meaningful,
OEMs will look to harness the power hands free and convenient new
of voice AI in their vehicles. The experience for consumers, and in
market potential for our offering is as turn adds more value for OEM
big as the automotive industry itself.” partners,” he adds.

October / November 2024 Software-Defined Vehicle Magazine 11

 SDV success
requires a
harmonised
approach
Accenture’s SDV-oriented LearnVantage Academy aims to
transform existing strategies for automotive software
development. By Stewart Burnett

12 Software-Defined Vehicle Magazine October / November 2024

 I
ndustry efforts to realise
software-defined vehicles (SDVs)
have proven far from smooth.
Various automakers have
reported their struggles to establish a
proper strategy—a lack of cohesion
and many disparate teams working
on separate aspects of the vehicle are
recurring themes. Even partial
software integration has faced
significant bottlenecks. “We have 150
different companies that develop
software [for us],” remarked Ford
Chief Executive Jim Farley on an
episode of the Fully Charged podcast. Juergen Reers, Senior Managing Director, Global
“They’re written in over 100 different Automotive and Mobility Lead at Accenture
languages that don’t talk to one
another, and we can’t even Since the automotive industry is still
understand it all.” This decentralised getting to grips with software
and multitudinous approach to development, realising the potential
design has hamstrung the speed at of SDVs may not be an easy task.
which core software is created and Therefore, digital transformation
updated. consultancy Accenture is working to
help automakers overcome this
In response, some automakers have teething stage by introducing an
made relatively drastic moves to ‘academy’ intended to provide the
shore up their developmental structural backbone for SDV
resources, sometimes to chaotic development. This academy places
effect. For example, Volkswagen talent transformation at the core of its
announced in July 2024 that it would vision of SDV realisation and eventual
invest up to US$4bn in pure electric adoption, guided by a “digital native”
vehicle (EV) brand Rivian as part of a mindset.
new joint venture to create “next
generation SDV platforms.”
Volkswagen Chief Executive Oliver Supporting all parties
Blume noted it would help lower the
cost of its software strategy, reducing Accenture emphasises the
its reliance on wholly-owned software importance of clear process and
firm Cariad, which struggled with smooth communication to the
overrun budgets and two-year delays. success of SDVs. “It’s not only what to
Amid threats of factory closures and do differently, but how to do things
mass redundancies, however, the differently,” explains Juergen Reers,
investment has been received poorly Senior Managing Director, Global
by staff. Works council Chair Daniela Automotive and Mobility Lead at
Cavallo characterised the failures of Accenture. “This is what allows you
Cariad as the result of “egoism” to bring new features at speed and
between Volkswagen’s competing scale in a manner that’s very different
brands. “Can we be sure this [joint to the delays you see today in
venture] won’t be the next billion- software programmes.” Such
euro grave?” she asked. struggles are the reason there is a

October / November 2024 Software-Defined Vehicle Magazine 13

 timely and widespread need for a adapt learning systems to meet the
learning platform specifically geared unique needs of both individual
towards SDVs. teams and team members, ensuring
that every stakeholder has exactly
Accenture’s LearnVantage academy what they need to thrive in a newly
for SDVs was established to help its SDV-oriented environment. “This is a
automotive clients’ workforces meet huge transformation for the industry.
the needs of intensive software Education is an important
development. It aims to educate component, and it begins with
technical teams, C-suites, and board developing talent,” remarks Reers.
members on how to succeed within
this new framework. It will offer
technology learning programmes; A practitioner’s
specialised, predesigned technology approach
academies; ecosystem learning
certification services; managed One of the most important aspects of
services for a client’s own learning this “talent transformation” will be the
capabilities; and nanodegree promotion of a pro-digital mindset.
programmes, certified online “Digital nativism is the ‘big ask’ in the
programmes designed to provide automotive industry today across all
users with hands-on experience and functions, and the absence of this
industry-relevant skills in specialised mindset is one of the key impediments
fields. All courses will be highly to most automakers’ transformation,”
personalized with artificial explains Raghavendra Kulkarni,
intelligence (AI) to meet the specific Managing Director for SDVs at
needs of teams and individual users. Accenture. As he explains, it is
“foundational” for automakers to have
In March 2024, the company talent that exemplifies this mindset
committed to investing US$1bn over both in decision-making positions and

“ It’s not really about teaching

particular knowledge sets; it’s
about delivering a practitioner’s
approach to SDV development

three years into the academy, throughout core teams to mitigate the
including the acquisition of digital need for expensive acquisitions and
learning platform Udacity. The continuous talent hunts. These teams
platform uses AI to continuously must work harmoniously towards

14 Software-Defined Vehicle Magazine October / November 2024

 “ We have 150 different
companies that
develop software [for
us]. They’re written in
over 100 different
languages that don’t
talk to one another,
and we can’t even
understand it all.

shared goals with open affordably in the absence of

communication in place and minimal congestion and irrespective of
space permitted for divergences in whether they have a driver’s licence,”
approach. he states, largely in reference to
autonomous driving. “But more
Automakers, Kulkarni continues, broadly, digitisation, software
cannot become software companies platforms, and digital services offer
in spirit alone: it is a shift that must huge opportunities to accelerate this
touch upon every aspect of a transformation.”
company’s practice, including who it
employs, the structure of its Accenture wants to situate itself as a
workflows, and how its employees “transformation partner” for
approach every design challenge. “It’s automakers on this journey.
not really about teaching particular Ultimately, what differentiates the
knowledge sets; it’s about delivering a LearnVantage Academy from more
practitioner’s approach to SDV standardised educational
development via this academy.” programmes is that it helps break
down the walls between different
The relevance of Accenture’s learning aspects of software development. “It
platform extends beyond the is all about integrating the digital
realisation of SDVs for their own sake. strategies of customer experience,
For Reers, it will reflect his vision for product digitisation and enterprise-
a more sustainable, carbon-friendly wide digital solutions,” concludes
future mobility sector. “It’s about Kulkarni. “Stitching all of these things
offering the freedom to move both together into a unified vision is what
goods and people, not only without the industry needs to move forward
carbon emissions but also safely and with SDVs.”

October / November 2024 Software-Defined Vehicle Magazine 15

 An open-source
OS changes the
value proposition
of vehicles
Elektrobit’s open-source OS facilitator, EB corbus
for Linux, could help usher in a new era of residual
value opportunities for SDVs. By Will Girling

S
oftware-defined vehicles Operating system (OS) innovation will
(SDVs) are increasingly become an essential for creating the
expected to include a range vehicles of the future. The global
of advanced safety, driver automotive OS market was valued at
assistance, self-driving, and US$11.83bn in 2023, according to
infotainment features. Importantly Grand View Research, rising to
for electric vehicles, all of these around US$30bn by 2030. To date,
functions will need to work this growing demand has been
harmoniously while consuming as fulfilled through a multitude of
little power as possible. More than closed-source proprietary solutions
ever, automakers need the skills and specifically tailored for the
knowledge to act like tech automotive industry. This may not
companies. continue for much longer.

16 Software-Defined Vehicle Magazine October / November 2024

 High-level architecture

“When you factor in safety and means as many developers as possible

compliance rules for automotive- must be empowered to take part.
specific OS, both the entry barrier and “This requires an open-source OS,
cost for developers is really high, and and perhaps the most established is
turnaround times for new updates are Linux, the developer’s OS of choice.”
long,” states Moritz Neukirchner,
Senior Director of Strategic Product However, Linux’s huge global
Management, SDVs, at automotive community does not strictly adhere
software company Elektrobit. If to automotive development
automakers want to eventually iterate processes and generates around 500
their cars at a comparable speed to kernal updates—a programme that
smartphones, they will need to prevent conflicts between different
change their approach. computer processes—each year.
“That makes it impossible to quality
its safety for use in vehicles,” states
Bridging the gap Neukirchner. It took Elektrobit
around seven years to resolve these
Neukirchner tells Automotive World issues. Finally, in April 2024, the
that SDVs are motivating a company announced EB corbos
consolidation in the automotive OS Linux for Safety Applications—the
market. The need for constant first open-source OS that meets
software innovation that also meets automotive safety compliance
safety and compliance regulations standards.

October / November 2024 Software-Defined Vehicle Magazine 17

 The solution has been positively plan. According to this idea, vehicles
assessed according to ISO 26262 ASIL could actually become more valuable
B and IEC 61508 SIL 2 standards. “We over time. “It will be equivalent to the
run an off-the-shelf Linux but make it leap from regular cell phones to
usable in an automotive context,” he smartphones and the App Store.”
explains. A hypervisor enables
multiple virtual machines to operate The sensor hardware increasingly
on a single piece of hardware, so incorporated in vehicles for a range
although the OS can now be used in of connected and autonomous
high-performance computer domains functions offer a rich seam of
relating to features like driver- untapped potential. For example,
assistance or autonomous driving, it while some automakers might
is supervised to prevent interference initially want to include biometric
with safety-related functions. “This sensors for secure payments inside
means we can participate in the the vehicle, the open-source OS
development community without developer community might expand
infringing on industry requirements.” their functionality to include speed
In effect, EB corbos helps bridge the restriction for younger drivers. LiDAR
gap between automotive and tech. and parking sensors could also be
But what does this mean for SDV used for highly accurate weather and
development? road condition monitoring
programmes. “Without changing any
of the hardware, companies can add
Updateability value purely through software,” says
becomes value Neukirchner.

For Neukirchner, the promise of SDVs In much the same way that removing
is an inversion of how vehicles are the ability to update smartphone
typically valued. “When you drove a software would ruin the market, he
car away from the dealership, that believes new SDV ecosystems will
was the best it was ever going to be; become inextricable from the future
everything went downhill from of automotive. “Cars might remain
there.” By opening up programming mobility devices, but updateability
interfaces and development will become the value proposition.”
environments, he believes SDVs can Elektrobit is positioning itself to be
enable innovation that wouldn’t have the facilitator of this industry
been conceivable in the hardware inflection point through streamlined
manufacturer’s original business software development with Linux.

18 Software-Defined Vehicle Magazine October / November 2024

 “ Proprietary OS solutions [. . .]
cannot compete with the speed,
efficacy, and affordability of
open-source OS

Creating a developer Systems Alliance (COVESA) and

Eclipse SDV—that prioritise
community collaborative, community-based
solutions. However, the industry will
The importance of pursuing an open- need to address the current dearth of
source OS approach, Neukirchner auto-specific IT qualifications and
continues, is that creating a de facto skills holding back this movement’s
standard for vehicle software will full potential.
necessarily be a joint effort; no
automaker can achieve the The wider importance of expanding
portability required by itself. the software development and testing
“Harmonising some aspects of ecosystem is already clear. In early
software across every automaker and 2024, Porsche was forced to
creating a less fragmented market will discontinue the Macan in Europe, as
be a big future trend.” This is not to UN Regulation No. 155 (R155) cyber
say that individual brands will lose security compliance would have been
the opportunity for differentiation: prohibitively expensive. Proprietary
ADAS functionality, for example, OS solutions, states Neukirchner,
could vary according to each OEM’s cannot compete with the speed,
investment. However, he states that efficacy, and affordability of open-
establishing an open language and source OS. By bringing together
framework for development will developer communities and
accelerate updates. automotive standards, he concludes
that EB corbus could help prevent
Elektrobit estimates that EB corbus future product recalls on the basis of
can shorten product development cyber security and safety. The
cycles by up to 50% and subsequently ultimate result will be SDV systems
save significant R&D costs. Looking that are more durable, remain
ahead, Neukirchner envisages the market-viable for longer, and
solution joining a tapestry of open- subsequently create a fresh business
source automotive hard/software paradigm for the automotive industry
initiatives—such as Connected Vehicle based on new residual value.

October / November 2024 Software-Defined Vehicle Magazine 19

 Software-defined
vehicle advances
to AI-defined
automaker
Empowered by advancements in digital twins, automakers
are deploying AI-powered solutions to boost efficiency and
innovation. By Pedro Moreno

T
hree decades ago, the and the impact that new applications
introduction of electronics and services would have on their
into vehicles marked a businesses. Capgemini estimates that
significant shift in the 16% of large automotive OEMs can
automotive industry, requiring new increase operating profits by
skills and working methods. The deploying AI at scale.
arrival of active safety functions,
automated driving features, and digital AI will not only improve functions
cockpits accelerated this trend. The within the vehicle but will also
advent of over-the-air (OTA) updates enhance efficiencies across the
ushered in a new era, allowing automotive production workflow. That
vehicles to evolve after market release, includes anything from designers
integrating new functions and services using AI-generated images to imagine
customers would pay for. This is the new vehicles, to design and
era of the software-defined vehicle, engineering teams using generative AI
where software activities are primarily application programming interfaces to
limited to functions within the vehicle. connect their tools to build digital
twins of their facilities, to marketing
The landscape further evolved with and retail sales departments adopting
OpenAI’s release of ChatGPT in 2023. generative AI tools to brainstorm and
This development made industries develop marketing copy and
realise the potential of generative AI advertising campaigns.

20 Software-Defined Vehicle Magazine October / November 2024

 Most automakers are evaluating 100
to 150 use cases for AI deployment,
making it hard to imagine a single
function that will not be improved
with generative AI. This
transformation leads to a new era,
the “AI-defined automaker.”

Knowledge chatbots At GTC 2024, Ford presented a deep dive into

accelerating automotive workflows with Large
enhancing customer Language Models

service
Ford, which receives around 15,000 Automated vehicles
customer calls a day, uses generative (AV) 2.0
AI throughout the call centre
workflow. First, a speech-to-text AV 1.0, the traditional way of
system automatically manages the developing automated driving
data associated with the customer functions, is based on a modular
request while removing any architecture of perception,
personal identification information, prediction, and actuation. Perception
which is mandatory in many is primarily based on Convolutional
countries. This intelligent data Neural Network (CNN) classification
management enables the creation of AI, and it was a common belief that
a second AI service, a chatbot used by solving perception, meaning
by supervisors and engineers to having perfect classification of objects
query the database on questions in a driving environment, the
such as the number of call centre autonomous driving problem, even
calls or even how many were related for high levels of automation, would
to a specific issue. Additionally, AI be solved. However, after more than a
assists the call centre operator by decade of efforts, it became clear that
suggesting the best answers to the long tail of use cases is endless,
customer queries, thereby and more than perfect perception, a
improving call centre speed, right interpretation of the scene is
efficiency, and customer required for the vehicle to manage
satisfaction. previously unseen scenarios.

Another example is the use of Transformer engines, which are the

knowledge chatbots at repair shops foundation of generative AI, are not
and dealerships. For example, if a trained exclusively on automotive
vehicle’s air conditioning is broken, data but on general data, enabling
the technician can simply ask for the them to abstract and correctly
best way to repair it on the specific understand driving scenes that have
car model, and the chatbot will never been encountered before,
provide step-by-step instructions making the right decisions. This is the
and visual examples. core of AV 2.0, where an end-to-end
architecture, with sensors’ raw data

October / November 2024 Software-Defined Vehicle Magazine 21

 Debrecen is the BMW Group’s first facility to be planned and validated
completely virtually thanks to a partnership with Nvidia

as input and vehicle actuation Digital twins enable physically

(acceleration and direction) as output, accurate visualisation and precise
replaces the traditional AV 1.0 modular testing of different product versions
architecture. In addition to and scenarios. They allow for
performance benefits, this approach superpowers: travel in space to
optimises the computational resources virtual factories or inside vehicle
allocation, as there is only one prototypes to help teams optimise
function being calculated, rather than designs, and travel through time to
several modules competing for understand past errors and avoid
available resources. It also facilitates future ones by working on alternative
the transfer of a vehicle to different futures. Instead of teams relying on
cities and countries, a difficult problem in-person meetings and static
for rule-based AV 1.0 systems. planning documents for project
alignment, digital twins streamline
Startups like Wayve, a developer of communication and ensure that
embodied AI for assisted and critical design decisions are informed
automated driving, and Waabi, an AI by the most current data. Digital
company building the next generation twins allow project stakeholders to
of self-driving technology, are making visualise designs in context during
tremendous progress in this field, design. Teams can identify errors and
demonstrating impressive results. incorporate feedback early in the
review process.

The superpower of Recent advancements in digital twins

digital twins include the adoption of OpenUSD
(Universal Scene Description), an
Digital twins are physically accurate extensible framework and ecosystem
virtual replicas of assets, processes, for describing, composing,
or environments that exist in the simulating, and collaboratively
physical world. In industry, digital navigating and constructing 3D
twins can be used throughout the scenes. HTML is the standard
product lifecycle, from design and language for the 2D web, while USD
manufacturing to maintenance and can be seen as the powerful,
operations, to sales and marketing. extensible, and open language for the

22 Software-Defined Vehicle Magazine October / November 2024

 3D web, enabling communication teams and customers the ability to
between the many different tools and interact with assets and knowledge
siloed departments in the product life bases in natural language, navigating
cycle. The latest developments in the digital-twin environment, and
physically-based rendering and creating visually and physically
accelerated scalable computing accurate digital worlds for product
enable the creation of large-scale, development and training, such as
physics-based digital twins that are teaching autonomous vehicles to
truly representative of the real world. drive or robots to walk. The digital
twin becomes a digital gym or
In early 2022, BMW introduced the training ground in which the
iFACTORY for virtual planning in product’s software functions can be
digital twins. All of the BMW developed. Digital twins are also used
Group’s vehicle and engine plants for synthetic data generation,
were 3D-scanned, as well as more creating the data required to train the
than seven million square meters of AI algorithms that will be used by the
indoor and 15 million square meters product being developed.
of outdoor production space. BMW
has a vast workforce of factory
planners, including those working at The AI-defined
facilities like the future Debrecen automaker is here
plant, who play a critical role in the
planning and operations of the today
company’s factories worldwide.
Their jobs are highly complex, and The automotive industry is
even the slightest miscalculations or undergoing a profound
mistakes can result in massive real- transformation, not only introducing
world costs. software-defined vehicles, but also
entering the era of the AI-defined
This is why physically accurate virtual automaker. Generative AI is
planning is highly desirable, as it revolutionizing every aspect of the
allows planners to experiment and business, including design,
make changes at practically no cost. engineering, customer service and
Planners can pre-optimise autonomous driving. Empowered by
production processes using virtual advancements in digital twins,
environments before committing to automakers are rapidly deploying AI-
massive construction projects and powered solutions to boost efficiency,
capital expenditures. This approach innovation, and customer
significantly reduces costs and experience. As the industry embraces
production downtime caused by this AI-driven future, every function
change orders and flow re- can be enhanced and optimised by
optimisations on existing facilities. the power of generative AI. This
evolution promises to unlock new
The combination of digital twins and levels of efficiency, innovation, and
generative AI further enhances the customer-centricity, positioning the
benefits of each. Generative AI can automotive industry for continued
assist in generating assets, providing success in the years to come.

About the author: Pedro Moreno Lahore is Senior Account Executive, Automotive, at Nvidia

October / November 2024 Software-Defined Vehicle Magazine 23

 May Mobility: AVs must
prioritise safety and
accessibility
To win over an increasingly distrusting public, AV
technology must recast itself as a pillar of local
communities and not a threat. By Stewart Burnett

24 Software-Defined Vehicle Magazine October / November 2024

 W
inning over an firm May Mobility, this means two
uncertain public is one things. First, developing self-driving
of the biggest software should not be bound to a
challenges facing the rigid and therefore potentially
realisation of autonomous vehicles dangerous rules-based system. An
(AVs). In the last few years, a wave of entirely rules-based system not coded
high-profile accidents has caused for every possible scenario will
public excitement for driver inevitably lead to unsafe behaviour.
assistance and self-driving technology Second, self-driving technology must
to turn into discomfort. be deployed in situations where
everyday members of the public—not
One notorious incident involved a a self-selecting audience of robotaxi
Tesla Model S using a beta version of users—can actually experience it.
the automaker’s Full Self-Driving
feature: the car came to an abrupt
halt while crossing the San Francisco Safety is everything
Bay Bridge, resulting in an eight-car
pile up. Nine people were injured, Unlike many firms in the autonomous
and traffic was held up for more than driving space, May Mobility has kept a
an hour. SAE Level 4 developers are low profile. This is partly due to a lack
faring no better. In November 2023, of controversy: in the nearly seven
General Motors’ Cruise had its licence years that May Mobility has been in
to operate on Californian roads operation, there have been minimal
revoked following an incident in San reported crashes, injuries, or traffic
Francisco the month prior. jams. Despite this, the company has
continued to raise significant amounts
Consumer sentiment towards of money while others have seen their
automation technology is nearing funding dry up. In November 2023, it
rock bottom in the US. 93% of announced a US$105m Series D
respondents in a January 2024 Forbes investment round led by Japanese
Advisor poll indicated they had telecommunications company NTT.
concerns about some aspect of self-
driving cars, with safety and The secret to this success is May’s
technology malfunctions the most prioritisation and strong track record
widespread. 61% indicated they of vehicle safety. “We develop AVs,
would not trust an AV to transport a but we’re particularly focused on
loved one. These concerns are validating this amazing technology
backed by National Highway Traffic that just works and is safe to use,”
Safety Administration data: 9.1 remarks Sarah Gryniewicz, Strategic
crashes occur per million driverless Commercial Development Lead at
miles, compared to only 4.2 in May Mobility. Her dream for
human-operated vehicles. autonomous driving is simple: it must
not stand out to the vehicle
Growing distrust means that occupants. Rather, they should
companies operating in the almost forget they are experiencing
autonomous driving space must make an AV. “It should be boring because it
significant efforts to demonstrate the is driving completely safely—in fact,
safety and reliability of their boring is a really fantastic goal to
technology. For Michigan-based AV work towards.

October / November 2024 Software-Defined Vehicle Magazine 25

 May Mobility’s engineers blend emergent behaviour with
foundational knowledge such as local road maps

Creating a “boring” autonomous Such intensive simulation may give

driving experience means driving in the impression that May Mobility’s
an entirely predictable and human- self-driving software is particularly
like manner, enabled by May demanding and can only run on the
Mobility’s patented self-driving most advanced hardware, but
software algorithm, Multi-Policy Gryniewicz dismisses this idea.
Decision Making (MPDM) technology. “MPDM is designed to home in on
Instead of an AV adhering to a rigid scalability—it’s extremely energy and
and rules-based system, which can data-efficient.” She explains that the
lead to serious issues when particular method of decision-making
encountering a situation for which it uses less compute power than relying
has not been prepared, MPDM is on a large base of existing data. This
developed to allow for emergent isn’t to say the system entirely
behaviour and enables the car to disregards the benefits of
navigate through almost any situation. foundational knowledge: before a
vehicle is brought into operation in a
MPDM works by continuously town or municipality, it is equipped
running thousands of simulations on with and trained on a map of the
vehicle hardware in real-time to local area.
anticipate any possible change or
development in the driving situation.
Using this information, the vehicle is Bringing AVs into the
able to decide for itself what action is mainstream
the safest. “What makes our software
special is that it’s learning even while “We think AVs powered by our
it’s on the road,” adds Karsten software can make a difference
Kutterer, Senior Communications everywhere, not just in major cities
Manager at May Mobility. “When you and robotaxis,” explains Gryniewicz.
have a team of engineers, they can “There’s a ton of people out there for
only imagine so many rules to put whom there is a mobility gap, so
into the code.” having something that’s scalable

26 Software-Defined Vehicle Magazine October / November 2024

 “ We want to enact change within
the community that makes up
our customers and partners

makes it more useful across all of economics, and you can use AVs to
these applications.” She states that fulfil that need.”
May Mobility’s technology is designed
to be platform agnostic, able to be The company’s strategy is to start at a
tuned and adapted across a variety of small scale within a given area and
different vehicle types, ranging from then expand as trust grows. This will
passenger cars to buses. be achieved not only by demonstrating
to locals that the technology is safe to
Gryniewicz emphasises that the use but also by prioritising accessibility
company’s system enables a degree and expanding mobility access in areas
of affordability that other solutions that were previously lacking. “We want
may not be able to reach. Crucially, to enact change within the community
May Mobility achieved this without that makes up our customers and
sacrificing other pivotal redundancy partners,” says Gryniewicz.
features, such as backup power and Accessibility—such as for those in
sensor communication systems and a wheelchairs—is also a key AV design
fallback vehicle safety system. priority.

In combination, these factors are This approach forms not only May
intended to make May Mobility’s Mobility’s strategy for commercial
autonomous driving software, success but also what it believes is the
alongside the actual vehicles it best way to sway consumer sentiment
produces, the ideal solution for its towards the AVs. Rather than
target audience: transit agencies, presenting it as a threat to human
municipalities and businesses. “We’re jobs—a common critique of robotaxi
very bullish about these segments. firms—the technology is instead cast
There are transit companies all over as a tool for safely connecting
the US that need to fill gaps and help communities and individuals with
people get around. I think there are limited mobility options. “AVs need to
really good use cases for autonomy in serve all people, and we design them
that,” she explains. “These cities and for that purpose. That’s something
private companies have budgets, and we’re proud of, but it’s also something
we all know they are facing driver that’s especially important to our
shortages. They need sustainable markets,” concludes Gryniewicz.

October / November 2024 Software-Defined Vehicle Magazine 27

Luminar: software is
the key differentiator
in sensor tech
While the automotive industry debates what sensor
hardware is optimal for autonomous features, software
could be the key differentiator. By Will Girling

28 Software-Defined Vehicle Magazine October / November 2024

 D
emand for connected the data it generates,” comments
vehicles, advanced driver Jefferson. “OEMs often cycle
assistance systems (ADAS), alternately through wanting to create
and autonomous driving ADAS/AD software in-house and
(AD) are all prominent factors driving partnering with a third party. The
automotive software development truth is very few have invested in
today. The global market’s value is software tied to high-quality LiDAR
expected to reach US$117bn by 2032— data.” It was by identifying this gap
up 241% from 2023, according to in the market that Luminar arrived
Precedence Research. at the idea for Sentinel.

However, unlike in the tech industry, Development started based on

automotive players must carefully perception software used for the
balance the desire for a fast time-to- company’s LiDAR sensors: a point
market with the safety considerations cloud—collection of data points
of deploying new ADAS/AD features arranged in 3D space—which
on the road. “The software needs to facilitates the identification of
pull together radar, cameras, LiDAR, sidewalks, lane boundaries,
compute platforms, and other pedestrians, vehicles, etc. “From
hardware,” says Aaron Jefferson, Vice there, it was a case of deciding how to
President of Product at Luminar package it in a stack and match it to
Technologies. “The big challenge is the KPIs customers might want to
engineering software that works first track, such as accuracy of object
time out of the box.” detection in low levels of lighting.”
This took some time: an initial
More frequent industry discourse Sentinel prototype with limited
around software-defined vehicles capabilities emerged in 2021. A
(SDVs), he tells Automotive World, can commercially viable version was
give the false impression that this launched in July 2024.
challenge is now easily surmountable.
“In fact, it still takes a lot of design, The final product consists of a
development, and testing to make compute platform based on an Nvidia
sure software won’t glitch once it’s in Orin processor, a LiDAR sensor with
use.” As such, Luminar has been core software and point cloud
developing a new stack—Sentinel—to output, a perception stack, and 3D
help provide automakers with the mapping and localisation features.
toolkit necessary for safely realising This suite is then integrated with an
next-gen vehicle functionality. existing system. Sentinel’s capabilities
are diverse—or “à la carte” in
Jefferson’s words—as each
Assessing true value automaker’s initial architecture and
project goals will invariably differ
Although it’s a Tier 1 more from another’s. “Luminar wants to
commonly associated with LiDAR provide a ready-made kit so that
and machine perception technology, OEMs can develop the safety and
Luminar is pushing past a strict functionality of their software
focus on hardware. “The value of without having to invest a billion
LiDAR is ultimately what you do with dollars in their own,” he states.

October / November 2024 Software-Defined Vehicle Magazine 29

Luminar’s Sentinel stack combines a suite of features so that automakers
can scale the functionality of LiDAR in any direction they desire

production immediately. Rather, it is

Broadening tech complementary software for helping
horizons more mature players assess and
improve LiDAR-based features already
In April 2024, insurance firm Swis Re under development. Although it is
published an audit of Luminar’s still too early for the company to
LiDAR combined with a pre-release report on individual customer use
version of the Sentinel stack. In total, cases, these are currently ongoing.
800 individual tests—including car-to-
car and car-to-motorbike incidents Much in the same way that Luminar
and vulnerable road user scenarios— partnered closely with Volvo Cars to
were conducted on a vehicle with integrate LiDAR sensors seamlessly
SAE Level 2 ADAS functions. Swis Re into the roofline of its EX90, Jefferson
found that a vehicle equipped with believes an ongoing industry dialogue
Luminar’s technology was 25% less can help marry ADAS/AD software
likely to be involved in a collision with brands’ wider goals. “Based on
than an equivalent vehicle without it, our back and forth with an automaker
and overall collision severity was in the Asia-Pacific region, it’s already
reduced by up to 40%. become clear that new software
solutions can help manufacturers
While this research is promising, realise the unexplored potential of
Jefferson notes that Sentinel cannot their existing hardware.” It is by
be used simply as a plug-and-play capitalising on these broadened
shortcut for bringing the next horizons, he adds, that future SDV
generation of ADAS/AD solutions into innovation could be realised.

30 Software-Defined Vehicle Magazine October / November 2024

 “ We’re going to see [LiDAR’s]
democratisation across all vehicle
segments as software helps
deliver the sensor’s true value

beyond. “The challenge is that this

Democratising LiDAR will represent a paradigm shift from
Although the industry has not yet the camera and radar fusion with
reached consensus on what sensor which many OEMs are comfortable.”
hardware or fusion will enable mass However, he believes the rigours of a
market Level 4+ autonomy, many rapidly changing industry make
consider LiDAR’s functionality previous solutions inadequate.
valuable for building consumer trust
during a difficult time in the market. One of the most attractive prospects
High-profile incidents involving AD of SDVs is that they can unlock
systems have decelerated interest previously inaccessible monetisation
among stakeholders, but a software- opportunities. For the first time in
driven re-examination could help the automotive industry’s history,
development regain momentum. vehicles could become more
From an engineering perspective, valuable with time, not less, as new
Jefferson states, “we haven’t even features can be added quickly
scratched the surface of what LiDAR through software. For ADAS/AD,
can deliver.” automakers will need a system that
provides foundational safety for the
In April 2024, the US National mass market while also containing
Highway Traffic Safety Administration the scalability for advanced AD
stated that automatic emergency features. Software stacks based on
braking (AEB) must become LiDAR, concludes Jefferson, can
standard—rather than a luxury—in all help fulfil this need. “LiDAR is
cars and light trucks by 2029. fundamentally a safety technology,
Jefferson anticipates that LiDAR not just a facilitator of autonomy; it
technology expanded and improved works equally well for both. We’re
through software will play an going to see its democratisation
important role in fulfilling this across all vehicle segments as
requirement, setting the foundation software helps deliver the sensor’s
for Level 2 ADAS functionality and true value.”

October / November 2024 Software-Defined Vehicle Magazine 31

 DeepRoute.ai CEO:
SDVs and AGI
herald ‘the era of
robots’
DeepRoute.AI’s CEO believes software-defined
vehicles are the key to artificial general intelligence in the
physical world. By Megan Lampinen

32 Software-Defined Vehicle Magazine October / November 2024

 A
rtificial intelligence (AI) is energy vehicles came with integrated
shaping the development SAE Level 2 and L2+ functionality.
and functionality of
software-defined vehicles But as the industry moves towards
(SDVs). The promise is great but the greater levels of automation, another
journey has only just begun. Many form of AI is gaining traction:
industry players are working towards artificial general intelligence (AGI).
the vision of driverless cars that can While GenAI refers to algorithms that
navigate 24/7 in any and all generate new content—videos, code,
conditions. For Chinese self-driving images, etc.—AGI acts more like a
company DeepRoute.ai, the AI human in terms of common sense,
capabilities needed to realise that understanding and learning. It can
paradigm open up tremendous then apply that ‘general’ knowledge
potential in applications far beyond to all sorts of tasks. “The most
the roadways. important difference with AGI is the
generalisations,” Zhou tells
Automotive World.
The starting point
Today’s self-driving systems are
Led predominately by China, highly tailored and trained for
connected and intelligent interactions specific use cases and regions, usually
in vehicle cockpits are rapidly relying on a high-definition (HD) map
becoming the norm. Most Chinese that needs constant updating. But
consumers purchasing even a mid- with AGI, an autonomous vehicle (AV)
range car expect it to offer a host of that can drive in London can also
smart driving functions. “Today, cars drive in San Francisco or Beijing with
are controlled from screens and its generalised learnings applied from
preferably by voice interaction,” says one city to another. “Waymo can only
Maxwell Zhou, Chief Executive of drive in a few places like Phoenix or
DeepRoute.ai. “Automakers are San Francisco,” says Zhou. “If it goes
starting to integrate ChatGPT to somewhere else, it won’t work. The
improve the in-car digital assistants power of the new AI technologies is
and make interactions even smarter.” totally different and gives us the
Volkswagen Group and Stellantis are potential to drive everywhere.”
two big-name players leading the
integration of these early generative It also means there’s no need for an
AI (GenAI) systems across their vast HD map, which has been one of the
line-ups. unique selling points of
DeepRoute.ai’s self-driving system.
GenAI is also facilitating real-time “You would need to hire thousands of
decision-making in automated driving people just to maintain these maps,
systems, which are gradually taking and you would need to cover
on more of the driving tasks, everywhere—Europe, the Americas,
particularly in urban areas. “This is a China. It’s simply not going to be
top priority in new car purchases,” possible. AGI is the way to Level 5
Zhou points out. Data from the China autonomous driving, as well as to
Passenger Car Association shows that robots,” Zhou proclaims.
in 2023, more than 55.3% of new

October / November 2024 Software-Defined Vehicle Magazine 33

Cerence’s Chat Pro leverages a multitude
of sources, including ChatGPT

Data is the key more data and train your models,”

he says. “There’s a lot of work to be
Zhou, who has led autonomous done, but data is the key.”
driving projects at Baidu, Texas
Instruments, and DJI, suggests cars The learnings can feed into a
are the starting point for a wider foundation model that could be
evolution within all of robotics. easily transferred to other robots’
Specifically, they represent the first scenarios. And in the opinion of
kind of robots that will exist in the many industry players, cars are
tens of millions of units. Hedges and indeed becoming robots. In his 2024
Company estimates that there are GTC keynote, Nvidia Chief Executive
about 1.5 billion vehicles on the Jensen Huang asserted, “Everything
world’s roads today. Over time these that moves will be robotic—there is
vehicles will inevitably be retired no question about that. And one of
and replaced by highly automated or the largest industries will be
fully autonomous vehicles. These automotive.”
vehicles will produce enormous
amounts of data about the physical As Zhou explains, “The foundation
world, which can be harnessed to AI model that we train is based on
further train and iterate on the AI the data we collect from cars. It
algorithms. “You need to collect could benefit all robots. In the past,

34 Software-Defined Vehicle Magazine October / November 2024

 “ The power of the new AI
technologies is totally different
and gives us the potential to
drive everywhere

robots were built for a single launched a robotaxi service,

purpose, and that purpose would concentrated in the central business
need to be defined. But we’re districts of Shenzhen. It is currently
moving towards this new approach working with a Chinese automaker
in which there is no need to input a on mass production of smart driving
specific definition for the robot task. cars and at least three mass market
If these models work for car models are expected to debut
autonomous driving, they should later in 2024. As these and other
work for other robots.” similar systems appear in vehicles,
they can feed into the foundation
model, which can be migrated to
The starting point other forms of robotics thanks to the
move towards what Zhou calls ‘AI
One of the most important aspects 2.0’. As he emphasises: “We really
of training these AI models is the see the power of the new AI; it’s not
need to understand the physical like traditional AI. Up until last year
world. “There needs to be common we were trying to enter the data and
sense,” says Zhou. “The AI needs to train the models, but we realised we
understand distance, humans, how simply couldn’t solve the problem
vehicles work—for instance, that that way. Using this new architecture,
they don’t drive on top of a fence. we solved it. This new technology
We believe the common sense in should be able to migrate for all
these neural networks will robots. The era of robots is coming.”
eventually be transferable for other
tasks, and the best place to start is And the timeline? Zhou suggests that
with autonomous cars.” within the next five years the world
could see “a lot of general robots”
In 2021, DeepRoute.ai launched a across various applications. As for
production-ready autonomous the “era of robots”, that could be
driving solution that does not rely on another ten years, but he emphasises
HD maps. That same year it also that “it will definitely happen.”

October / November 2024 Software-Defined Vehicle Magazine 35

 SDVs will be enabled
by “deep integration”
of generative AI
Generative AI could shorten software development cycles
while also creating distinct, hyper-personalised customer
experiences. By Stewart Burnett

S
oftware-defined vehicles US$5bn into Rivian to develop an SDV
(SDVs) promise to transform platform.
the mobility landscape but
realising them is no easy task. Clearly, automakers need all the help
Even the launch of vehicles with they can get, and there are many
partial software integration has companies looking to pitch in
proven rocky. For example, Volvo solutions. Technology services firm
Cars stalled the launch of its EX90 Tata Elxsi—part of the wider Tata
SUV by almost a year due to problems Group—believes the deep integration
with software development. The of generative AI tools will be crucial
vehicle will finally ship in Q4 2024, for any automaker looking to be
albeit with fewer features than competitive within the emerging SDV
initially expected. For Volkswagen, space. Many automakers have already
software development has proven experimented with generative AI in
highly expensive. In addition to the areas such as vehicle architecture
US$13.3bn it has poured into wholly- prototyping. However, Tata Elxsi
owned software firm Cariad—only to wants to see the technology
be racked with numerous delays to implemented not only throughout
vehicle launches—it has also invested software design at the coding and

36 Software-Defined Vehicle Magazine October / November 2024

 “ Any application that is required
to interact with the vehicle
absolutely needs to be handled
offline

validation levels but also in a range of which are specifically targeted for SDV
customer-facing applications. vehicles. VFMs are core building
blocks for SDV co-pilots that can aid in
every process of the design lifecycle.
Establishing the
foundations It is crucial to supply the LLM with an
“enormous” amount of training data
“At Tata Elxsi, we are putting in a lot so that it can generate contextually
of effort to help our workforce and relevant and useful results. This may
automotive clients get ready for require some work beforehand to
generative AI and SDVs,” Biswajit convert data into compatible file
Biswas, Chief Data Scientist at Tata types. Some data—such as real-world
Elxsi, tells Automotive World. vehicle use, including the owner’s
Regardless of application, the journey sensitive personal information—will
towards generative AI integration require the careful navigation of
begins with laying out extensive regulations, especially if transported
groundwork: “This is an important across borders.
transformational tool, but it also
requires quite a lot of different pieces
to work effectively.” Transforming vehicle
design
The first step is to bring in large
language models (LLMs) alongside Once these fundamentals have been
other types of AI—typically under set in place, a wide variety of
lease from major tech companies—to development tools can be
conduct the fundamental task of implemented. “With the right tools,
generation. LLMs, for instance, can you can fine-tune generative AI
generate everything from written technology to make vehicle-specific
content to programming code. resources,” remarks Biswas. These can
However, all of these require a be developed internally or outsourced
significant amount of computing from other companies. Tata Elxsi, for
power to function. As such, Biswas example, offers tools for coaching
recommends also implementing autonomous driving, as well as an SDV
“Vehicle Foundation Models” (VFMs), development and validation

October / November 2024 Software-Defined Vehicle Magazine 37

Generative AI could potentially reduce
vehicle software development time

framework called Avenir, which existing software code and so may

automates the process of software not be a viable tool for companies
testing and interoperability, and just getting started on their journey
according to Khan, “can accelerate the towards SDV development. The more
SDV journey of OEMs and tier ones”. plentiful and diverse the training
data, Khan emphasises, the more
Given the recent struggles of useful the resulting code will be.
automakers to develop software in a While a human engineer is generally
timely manner, however, the most expected to intervene in this process
promising use of generative AI could by reviewing the resulting generated
be automatic code generation. While code, this could also be automated
this could be done via a text-based using LLMs. “The generative AI
prompt, Tata Elxsi believes it should frameworks can be utilised for review
be as intuitive as possible. “If we are of the source code, and then we can
using voice-enabled features for also use LLMs to generate new codes
customer-facing applications, we can within the given requirements.”
also use them to allow engineers to
interact with the systems they are
developing. This is far better than Better customer
developing everything from scratch,” experiences
states Jihas Khan, Practice Head,
Virtualization, Tata Elxsi. Looking to the future, Biswas believes
that generative AI could allow
He adds that this approach would consumers to actively participate in
save a “lot of cost and effort” for the design of their SDV, from physical
automakers and Tier 1 suppliers. At hardware and aesthetic features to
the same time, it relies on pre- the software capabilities it ships with.

38 Software-Defined Vehicle Magazine October / November 2024

 “ These cars are going to be data
centres on wheels, and that
means many more monetisation
options with generative AI

“People will have the option of using importance of automakers

an app to generate designs of their integrating 5G SIM capabilities into
car and select all the features they their vehicles at higher levels.
want. In this case, we’re also leaning However, even in this eventuality,
towards increasing the use of digital there will still be connectivity dead
twins.” This could be of particular zones. Biswas recommends that
interest to Gen Z, a demographic he automakers try to integrate as many
believes wants to actively shape and core generative AI features—such as
experience the products it uses the ability to relay information from
before they are physically produced. the user manual—into the vehicle
itself as possible. “Any application
Beyond vehicle design, generative AI that is required to interact with the
finds a range of applications within vehicle absolutely needs to be
the driving experience. The most handled offline.”
notable example is integration with
the voice assistant to enable Ultimately, the goal for automakers is
responses to nuanced or subjective to reduce development costs while
queries—for example, “What is the creating avenues for recurring
closest restaurant that offers food I revenue streams. Beyond voice
like?” However, Biswas is also a assistants, this could mean instantly
proponent of using the voice assistant customisable infotainment user
for applications beyond driver interfaces and the provision of highly
comfort. By integrating the vehicle’s tailored media content. Biswas
user manual into the LLM’s training concludes that generative AI will
data, the driver could troubleshoot in become invaluable for introducing
real-time. “You can check with your new, software-driven opportunities
copilot and ask about the nature of a even as the cost of physical vehicle
blinking light on the dashboard or the hardware— most notably electric
potential causes of a weird noise vehicle batteries—increases. “These
coming from the engine.” cars are going to be data centres on
wheels, and that means many more
One potential issue with generative monetisation options with generative
AI in customer-facing applications is AI. It is one of the most promising
the need for constant cloud ways to create new sales after that
connectivity. Khan notes the initial vehicle purchase.”

October / November 2024 Software-Defined Vehicle Magazine 39

 Automated ADAS
testing will take
autonomy into
the future
ADAS functionality is increasingly shaped by
software, and automakers will need thorough
testing methods to ensure their vehicles are safe.
By Will Girling

40 Software-Defined Vehicle Magazine October / November 2024

 A
dvanced driver assistance
systems (ADAS) of various
complexity are becoming
increasingly common in
the automotive industry, and this has
made thorough scrutiny imperative.
In the US, the New Car Assessment
Program (NCAP)—a consumer guide
that rates cars based on their safety
profile—has been testing ADAS since
2009 and continues to examine how
“Sometimes ADAS’ goal is to mitigate a collision rather
its metrics can be updated as the
than prevent one,” says Andrew Pick, Director of Track
technology itself evolves. Other Test Systems at AB Dynamics
regional NCAPs, including China and
Europe, are now similarly committed.

SAE Level 2 ADAS currently includes cannot be conducted on public roads

core safety functions like impending because of the inherent safety risk.
collision detection and prevention. Nevertheless, OEMs need “ground
However, some industry analysts truth data” upon which they can
contend that many of these partial validate system integrity. As such, AB
autonomy systems currently offer Dynamics manufactures test objects
limited to no real-world safety benefits. for proving-ground tracks based on
In fact, the US National Highway common real-world road hazards,
Traffic Safety Administration including pedestrians and
attributed 367 traffic collisions to ADAS motorcycles. However, in a rapidly
between July 2021 and March 2022. evolving market, this just provides
the baseline.
Meanwhile, software-defined vehicle
architectures enable automakers to “The main difficulty automakers face
expand autonomy beyond core safety is the sheer volume of tests,
to more advanced functions, combinations, and permutations they
including lane keeping assist and need work through,” says Pick. This
adaptive cruise control. The danger is applies to every single model
that widening the operating domain equipped with ADAS, and each
while even relatively simple functions regional market will have its own
are far from perfect risks alienating specific test requirements. To manage
industry stakeholders, whose trust this complexity, AB Dynamics has
has been repeatedly shaken by bad been developing an extensive
press. To rebuild and expand that protocol library in an attempt to
trust, reliable testing has never been simplify and automate what is
more important. otherwise a costly and time-
consuming process. A June 2024
update to its Track Applications Suite
Gathering ground added 237 tests for China NCAP, 60
truth data for the EU General Safety Regulation,
and 120 United Nations Economic
Andrew Pick, Director of Track Test Commission for Europe protocols.
Systems at AB Dynamics, tells “In total, our database now spans
Automotive World that ADAS testing more than 1,000 test cases.”

October / November 2024 Software-Defined Vehicle Magazine 41

 Thanks to artificial intelligence and says Pick Typically, this means
machine learning, sensors are measuring the time gap between a
becoming more discerning, meaning collision warning and the system
tests must be similarly sophisticated. taking action, such as applying the
Repeatability and realism, continues brakes, or recording the impact
Pick, are the crucial goals. The speed if a collision occurs.
protocol database factors in “Sometimes ADAS’ goal is to
information both about the vehicle mitigate a collision rather than
tested—for example, specific prevent one,” he clarifies.
dimensions or layout geometry that
could affect response—and the AB Dynamics claims that the
objects used. “If the customer can repeatability of automated ADAS
define exactly what they want, we testing using driving robots is much
can make it happen reliably on the superior to human testing. “People
proving ground. The realism comes just aren’t capable of the same
from the test objects themselves.” consistency: a driver could get the
AB Dynamics’ test objects have pathway, braking rate or speed right
identical radar signatures and during a test, but getting all three
comparable movement to what they perfect simultaneously is really
replicate: “That includes the motion tough,” Pick states. “A single
of pedestrians’ walking legs or successful test run could take six or
turning heads,” he explains. more attempts.” Comparatively,
robotic drivers are able to perform
strings of consecutive test protocols
The value of flawlessly and in a matter of minutes.
automated testing
During a test, vehicles are controlled Advancing ADAS with
by AB Dynamics’ range of driving software
robots—steering, pedal, and gear
change automation—instead of As vehicle autonomy advances
human drivers. An automaker’s beyond Level 2, the number of
selected test protocol then defines variables in a test scenario increase
the speed and relative positioning of significantly. In addition to
the vehicle and other objects during coordinating test objects and
the scenario at a coordinated ‘point providing a library of preconfigured
of interest’ on the track. This is test protocols, AB Dynamics’ Track
where the response of the ADAS Applications Suite includes a
feature in question will be assessed. Scenario Generator that can import
Aspects of the scenario can be previous test data and reshape it in
adapted by automakers in real time simulation into any conceivable
according to their requirements. scenario. Pick notes that this tool
can be used to establish a virtual
“Because we’re in complete control baseline for new ADAS functions in a
of the ecosystem, we can also safe digital environment. It also
manage the results by post- provides a data foundation for
processing test data to compare how validating those simulated results in
well ADAS performed against KPIs,” future real-world testing.

42 Software-Defined Vehicle Magazine October / November 2024

 AB Dynamics’ Track Applications Suite offers a comprehensive
portfolio of tools for testing ADAS safety automatically

While Level 4 autonomous driving play in proving to regulators and

systems exist in a limited capacity, consumers that this advancement
ADAS technology is far from does not come at the cost of safety.
reaching its peak in the mass Pick suggests that future ADAS test
market. According to scenarios could include objects in
MarketsandMarkets, global ADAS blind spots and upcoming hazards
hardware production is expected to obscured by blocked sensors.
almost double from 334 million
units per annum in 2024 to 655 Software-defined ADAS could unlock
million units in 2030. At the same billions of dollars in added value.
time, mainstream automakers like However, with a limited pool of
Mercedes-Benz and BMW are both talent from which they can draw,
exhibiting Level 3 systems—Drive automakers may need to build
Pilot and Personal Pilot, ecosystems that incorporate
respectively—and validating safety is software developers from Big Tech.
only going to become more While this could enable faster R&D,
complicated. Pick emphasises that these workers will need to be
the testing equipment industry must familiarised with the core
be prepared to move in step with parameters and data around which
these advances. they can iterate new ADAS features.
In this way, Pick concludes that the
“New ADAS functionalities are testing industry will be crucial for
increasingly defined by software,” bridging automotive and tech.
he observes. “As acceptance for “Software developers like to be
these technologies builds, the limit agile, and automakers want to
of their previous operating domain validate vehicles as quickly as
can be expanded.” Companies like possible. We can help them both
AB Dynamics will have a vital role to work in exactly the way they want.”

October / November 2024 Software-Defined Vehicle Magazine 43