Scaling Safety On The Journey To Automotive Autonomy - THE CONTEXT OF TRUST: veoneer.com
←
→
Page content transcription
If your browser does not render page correctly, please read the page content below
VEONEER
WHITE PAPER 2020
THE CONTEXT OF TRUST:
Scaling Safety On The Journey
To Automotive Autonomy
veoneer.com
We want to change
the conversation about
automated
We want to driving
change the commercial circumstances, such as shipping dedi-
cated truck platooning, taxis, or fully autonomous
across the fleet now, or many people may be left
out of the journey to automotive autonomy and
● Developing the best technologies, combined
with applying the data-driven insights that will
andconversation
safety about
autonomous driving
drones which may make small package deliveries
to homes. An even smaller fraction of vehicles will
fail to benefit from its innovations.
But there’s reason for hope, because a large
enable consumers to rely on every new safety
innovation, and therefore see them as core
be owned by consumers and operate on urban and number of those vehicles sold in 2030 will possess components of vehicle performance, and not
and safety rural roads, and will probably be high-end, costly advanced safety hardware and software. The key options, and
F
luxury models. will be to match those technologies to user needs,
During the journey to that future, people will make systems intuitive and so quickly. ● Provide carmakers with the platform to match
or too long, the public debate has focused
continue to be injured or die in driving-related technologies with driver use cases and car
on achieving full autonomy, a point at
types, from low to high-end models, and then
which a driverless car will operate on any
road, in any condition, as well as (or better)
accidents. Last year, the World Health Organization
cited road injuries as the 8th leading cause of
Improved safety can allow them to build on that framework by ea-
than a human driver. We are incredibly excited death worldwide: Traffic collisions resulted in no longer be merely sily adding (and swapping) new hardware and
software innovations.
by the promise of that future, and our engineers nearly 1.4 million fatalities, over half of which were
are helping lead the innovation and development pedestrians, cyclists, and motorcyclists, and 50 an option ● We believe that car users at all levels short of
necessary to get there. million people were injured. Fully autonomous full autonomy will drive for the foreseeable
We at Veoneer see the development challenge for
But the journey is far from over. Some forecasts vehicles will address only a small fraction of these future, not merely be driven around, and using
the next decade is not only to reach the goal of
project that only a fraction of cars on the road in incidents, whether powered by combustion or technology to augment their performance is
autonomous driving, but to democratize driver-
2030 — a decade from now — will be fully autonomous, electric motors. something we call collaborative driving. It is
assist safety technologies for the greatest societal
so it makes sense to talk about this transformation Further, here’s some sobering content: In 2013, now the time to accelerate in such technologies
impact along the way by:
not as a sudden revolution that will arrive someday, the WHO reported 1.25 million deaths related to that augment driver capabilities and thereby
but more like an evolution, or journey, that will unfold road traffic incidents, so the number has increased increase safety and enhance comfort in the
● Understanding the driving contexts that are
over several decades. by nearly 9% during the last 5 years in which we’ve more near-term timeframe, and to do so for the
relevant to the greatest number of people, such
The road to automotive autonomy isn’t a binary been talking about the safety benefits of smart or most vehicles in the largest number of circum-
as the qualities of their daily commutes to/from
leap from today to full autonomy. autonomous driving technologies. There are many stances…regardless of make, model, brand,
work, or the most common routes taken by
And what will 2030 look like when we arrive reasons this number has increased, most notably geography, or price tag.
drivers of rental cars.
there? If there are fully autonomous vehicles in because the total number of drivers has gone up,
operation they’ll most likely be utilized in controlled but it does indicate that safety features are needed
2 3
emerge when drivers are agitated, preoccupied, or to expectations. Forgiveness is a key attribute of
distracted by events inside the cabin (i.e. children) trust, and it’s imperative that consumers develop
or external factors not directly related to the road the faith to continue using a system or service even
(i.e. a flash of lightning or beautiful sunset). though it may fail sometimes. That continued use is
We believe the journey to autonomy would also required to give the system data from which it
benefit far more people if the industry focused on can learn and improve, so trust is also a core
defining and then providing the core safety tech- comment of continuous innovation.
PART I: nologies suited to such segmented use cases.
Trust drives the
The
Simply put, providing a buyer option for a package
of safety technologies unrelated to those specific
experiences — say, a “daily commuter package” or journey to
Democratization
“long-distance driver package” — is asking for a level
of consumer engagement (and expenditure) that
automation
doesn’t come across as a necessity. Currently, the collaboration between driver and
of Safety
Further, what types of safety technologies might system is basically a visual or haptic monologue
next be standard in certain models or vehicle types? during a short time window, not a dialogue within a
Establishing these benchmarks will help develop shared context of collaboration and experience.
the largest market opportunities and, therefore, the Today’s bells and buzzers are simply annoying
and capture attention without communicating
content, which can lead to distrust or simply
turning them off.
The democratization of safety will mean colla-
To bridge this gap, three qualities — driver condition borative driving that is a partnership – transforming
Context matters and circumstances, the need for technology to be drivers into users, as co-owners of the controls --
Nobody drives the same in every situation he or relevant and useful, and consumer perceptions of based on trust that both the vehicle and the driver
she might encounter in a given day, their driving ubiquity/reliability — need to merge to create what will have a hand in building based on their bespoke
habits can be impacted by a variety of internal we call a Context of Trust. It is within this context commuting or travel needs. Both people and their
characteristics, and every driver is different. that the application of driver-assist technologies vehicles will, quite literally, have to learn how to talk
The potential for new autonomous safety tech- needs to be understood, and to which future to one another.
nologies to fail to perform up to driver expectations innovation and development should be put. We see a new conversation about autonomous
in any of these all but infinite number of scenarios From a technology perspective, this means driving and safety that yields a near-term future in
means that consumers may have disappointing systems that go beyond indicators and warnings, which:
experiences and choose not to use those tools, or to tools that help drivers better manage their
simply never “opt-in” because of a priori mistrust. capabilities (i.e. attention/fatigue management) or ● Active Safety is no longer a premium feature
The industry knows from experience that there’s support understanding that encompass the design but is becoming standard in more mass market
no reason to believe that the introduction of more of system actions as well as system availability . vehicles
sensors or services alone will automatically address Commuting and long-distance driving are two greatest economies of scale and sales potential for ● Vehicles have more immersive, collaborative,
driver understanding and expectations. Many different safety contexts. carmakers. and trustworthy interfaces, making it more
continue to disable usage of intermittent vehicle This calls for matching safety tools to specific The goal should be to match the technologies to likely that drivers will use the safety tools at
automation such as lane keep assistance or and common uses. For instance, are lane keep the right use cases, then inform and engage with their disposal.
adaptive cruise control, citing their belief that the alerts an important safety mechanism for vehicles consumers on what those performance attributes
functions are unreliable, provide feedback at the used primary in congested urban settings? It’s might be and, finally, empower the automated ● Even though some vehicles may not have all
wrong times, or are simply annoying, according to certainly a crucial tool for long-distance highway and human actors in a vehicle to make reasonable the latest comfort features, they make use of
a study a few years ago. drivers. assumptions of the others’ performance and skill. trusted information gathered by other vehicles
Conversely, usage of anti-lock brakes (“ABS” Overlaid those environmental contexts are the Trust is the outcome of that process. It means — both similar and more advanced — allowing
systems) is just shy of 100%, insomuch that they various contextual states of driver state of mind. giving drivers the confidence in the necessity and them to benefit from the increasing body of
have been required technologies in all cars sold in People drive differently when they’re tired than suitability of the assist technology, and the capacity knowledge on safe driving.
the EU since 2004 and the US after 2013. when they’re wakeful and alert; differences also to assert control if it doesn’t consistently perform
4 5
● There is innovation on how vehicles function ● Were there conditions and/or situations in
and are marketed: For instance, cars’ driving which an existing or planned driver assist tech-
assistance systems could be customized to nology could have been applied, and what are
driving conditions (so city driving is the performance profiles of those use cases?
automatically supported by standard ADAS
with driver monitoring, but once on a longer We at Veoneer see an opportunity to capture such
distance highway the system switches on data and use it to continually innovate and improve
adaptive cruise control or a hands-off the way you and your vehicle collaborate (i.e. enable
function). Now, imagine paying for those you and your vehicle to learn). It can then be applied
features per use or even mile, much like broadly to other vehicles, and their learnings used
subscribing to a streaming service. to improve your safety profile, resulting in improved
collaboration between drivers and vehicles that
There is a need for a democratization of safety evolves over time.
technology and equipping all vehicles, from the Such collaboration will enable better safety
mass-market trim level to the premium level, with technologies and greater use of them, thereby
compatible safety systems that can learn and
improve as the industry innovates on its journey to
saving lives along our journey to automotive
autonomy.
” Vehicles need not
autonomous driving.
We at Veoneer not only intend to encourage
It is a challenging remit, yet a necessity, for at
least two reasons:
only to learn from
this new conversation but help lead its realization. First, as MIT’s Advanced Vehicle Technology
Research shows, consumers lack the core mental
other vehicles, the
A collaborative model of designers and a myriad of conditional
operating conditions don’t offer the support needed
infrastructure and
safety learning for successful technology adoption. For trust, one the driver, but explain
system of the pillars of collaboration, predictability is key.
Predictability can be improved when there is com- things in a clear way
to the driver. ”
For as much as today’s vehicles have in common mon ground between one vehicle and the next, and
with cars and trucks built five or more years ago, a common thought process running between them.
they are also very different. Vehicles have a greater People trust each other after they understand their
number of quality sensors, and the functional quality own and their counterparts’ limitations and skill sets.
of those devices continue to improve. The collection HMI requirements on system decision making
of driving data has also continued apace, created vast is a pivotal input to a collaborative driving scheme,
reservoirs of information that can be aggregated, making the ways safety updates are communicated
analyzed, and then applied to the function of both to drivers increasingly important. Mode error
existing and new vehicles. needs to be minimized, and the introduction of
Just think about it: The miles of road you and a system update that changes how a system is
others drive are experiences that improve your perceived needs to be explained in a way that
driving skills. Your vehicle can also learn from where drivers are ready to incorporate. Driver coaching
you drive and how you perform together, such as: will only become more important with increasing
automation.
● In what circumstances did you rely on a driver
assist technology, and how did it function?
● When did you switch off an assist technology,
and what were the conditions (which may lead
to insights as to why)?
6 7
Usability of safety
PART II:
technology will be as
important as its Scalable
availability Safety
Second, in order to build up and then apply the
amount of information needed for realizing this new
vision for safety, sensor data from all vehicles need
to be leveraged, and the resulting data merged
with other sources of information on the localized
context.
This will be accomplished by securely feeding
relevant data to the cloud, where risks can be
understood and profiled. Such risks can be related Applying safety capabilities It starts with a
to driver behavior, like understanding individuals’
tolerances for car-following and, thereupon, will be as important as modular system
determining based on vehicle dynamics the opti- innovating new ones architecture
mum timing for driver warning (or not). Other risks
can be related to security, where abnormalities As technology progresses, there is no reason to Cars already have the same bumpers regardless of
can be detected in the drive and the consumer be doubt that tomorrow’s low-end vehicles will have trim levels, and there is nothing to stop this from
alerted. Think less loud alarm or other warning, and more sensor technology than they do today. We also being the case for the rest of the electrical
more of a suggestion or nudge from a co-pilot. already see cameras and radars becoming standard architecture, yielding hardware as well as software
The result is that no driver would ever encounter in Europe and elsewhere, with a trend toward map- systems that are modular. Already, several OEMs HIGH
a road condition or situation that hadn’t been seen, ping and driver monitoring continuing to be strong. are taking this approach to developing the next
analyzed, and prepared for prior. No driver will ever Such a new basic level of fittings in every vehicle generation electrical vehicle architectures. The
have to drive alone. sold will present an opportunity for the vehicle to modular concept would allow for the bundling of
Information sourcing becomes a critical perceive its surroundings, itself and its occupants. safety and convenience features into three main
element in meeting this challenge as, contrary We see in this evolution the possibility for trim levels.
to what may be intuitive, vehicles with the least unlocking a whole new level of collaborative driving
amount of driver assist technology will account for and traffic safety by: For instance: MID
more than 50% of that data by 2025. Further, with
almost every vehicle having a camera (and/or radar) 1. Enabling vehicles to learn from each other ● The basic trim level ”BASE”, accounting for
somewhere between 2025-2030 in at least Europe and apply data insights to the various 60% of the vehicle volumes, could be centered
and North America, mass market vehicles will cap- configurations of hardware present in each around safety
ture the greatest amount of safety-relevant data. hardware platform, ranging from mass
● A MID trim level offer of advanced convenience
These vehicles also represent the greatest market to premium, and
and assistance features, such as adaptive
opportunity to apply that learning to the benefit
2. Establishing standards and a scalable platform cruise control or traffic jam assist, and BASE
of drivers, passengers, and pedestrians, even as
that allows carmakers to achieve synergies
the data are also useful to educate more advanced ● The HIGH trim level, accounting for only 5-10%
through the development of all trim levels due
systems (especially to determine relevant objects of total vehicle sales volumes, would aim to
to commonalities in software and hardware.
and the most probable paths in complex junction offer enhanced functions that are also
scenarios). We call this approach Scalable Safety. suited for higher levels of automation.
8 9
” We need a scalable architecture, where the
lower levels of vehicle systems are modular
and easily extended so that each level does
not require a whole new way of thinking about
hardware or software architecture.”
The goal would be to achieve a consistent safety learns. This also means we need to move further safety, as it’s primarily concerned with technical features on the camera differ regarding resolution
standard across the vehicle fleet, with systems along with standardization in terms of processing and economical synergies by designing one requirements, functional safety requirements or
allowing for updates of hardware or software as and interfaces, which is work that is going on today. common architecture instead of three (optimized) the required compute resources to process the
needed by OEMs. The challenge is to have a com- We need a scalable architecture, where the individual designs. image. Now, in order to satisfy the design criteria of
mon architecture design which allows expansion lower levels of vehicle systems are modular and A second important aspect is that sensor a scalable architecture described earlier, the same
from BASE to HIGH by adding hardware and easily extended so that each level does not require and compute platform design follows a systems scalability needs to be designed into the camera
software to the system while carrying on content a whole new way of thinking about hardware or engineering driven approach in order to offer the meaning common imager technology, common
from lower trim levels. software architecture. The lower levels can be built best (not optimum) technical solution for each of image processing, common interface design and
When OEMs have tested a system for their on the same basic system. For higher levels, there the system trim levels while keeping common common design for functional safety. Only then
volume level vehicles, those tests will still be valid are no current standards but still a competition on elements across them. can a system be expanded from BASE to MID by
for the higher trim vehicles. With a scalable archi- how this is to be achieved. As we scale across, and If you think about the forward-facing camera, replacing the BASE camera by a MID camera which
tecture, the same model can also be sold in several learn from the lower levels, we will build on these it is a key sensor in these systems and an enabler will still support the BASE content and allow field
trim levels, so that the end user can trim up or down learnings to agree on a working high-level archi- of the majority of all relevant safety features. The data collected with the BASE camera to be used by
their vehicle as they require. tecture that works for regular consumer vehicles. camera is used across all three trim levels from the MID system and so on.
Just as likely, OEMs could easily upgrade the BASE to HIGH, where it delivers information about The challenge and possible limitations in a
compute platform for their next model facelift, objects it detects in the dynamic driving scene scalable system design are imposed by the different
and thereby use added functionality as a tool to Scalable sensors which in turn are used for different features – trim levels as they serve different markets. BASE is
encourage trade-ins and/or new sales.
Regardless of trim level, the electrical archi-
put learning to work starting from a traffic sign information feature over
to an autonomous emergency brake all the way up
becoming the standard fitment, thus requiring very
cost-efficient solutions whereas HIGH demands
tecture will need to allow for updates from the OEM, A modular system architecture is only one aspect to an automated Highway Pilot feature. performance for convenience features while cost
reflecting the accumulated insights as the fleet of the Scalable Safety approach to democratizing As you might imagine, requirements from those might not be as important.
10 11
” With a common
architecture and modular
system development,
vehicles should more easily
learn from each other...”
Update, update, and then What does a scalable
update again architecture look like?
The scope of the learning challenge is broader than To implement such continuous improvement, ● Ability to design complete ADAS/AD sensing ● A diverse product range.
what’s contained inside the vehicle and must extend however, new tools for communicating need to be & compute architectures top-down.
- System components, designed to fit into
to understanding changes in the ground traffic designed to connect drivers with other vehicles
- System product planning identifying major scalable system solutions
system as a whole. As different parts of the vehicle (V2V), and vehicles and infrastructure (V2X). It’s not
market segments and common Safety/ADAS
eco-system update at different speeds, there is a only about the human-machine interaction inside - Those system solutions being complete,
content bundles and assumed use cases, which
need for understanding the modular sensing and the vehicle. System scalability not only will need offering all HW and SW as well as competency as
provide the baseline for system and feature
system setups, as well as when to issue various to incorporate the differences between ADAS and a System partner, including system integration,
development
updates to answer to consumer expectations of driver assistance, but also the ability to learn from V&V, functional safety, and so on.
continuous engineering throughout the vehicle the driver, from the other vehicles, and from itself. - “Real-life” research focused on traffic
lifecycle. In a software enabled system, vehicle safety accidentology and human factors driving
With a common architecture and modular systems need to evolve even better than the OS innovative and consumer centric solutions
system development, vehicles should more easily and applications on a new smartphone. When it’s - Scalable reference system architectures to
learn from each other; for instance, a vehicle with time for a software update, it needs to happen address different market segments and needs
more types of sensors can help map a trusted as expected and explained, with the advances in
route, which the basic vehicles can then use as functionality intuitive for users across diverse
input to their algorithms. technological backgrounds.
12 13
” As human drivers will
continue to play active
roles in driving their
The imperative for a vehicles while managing
new conversation about
autonomous driving
an increasing array of
and safety new or newly configured
The democratization of vehicle safety technologies
does not rest on an abundance of self-driving
technologies at their
vehicles. The aviation industry learned over 30
years ago that a philosophy of ’replacing’ the
disposal, they can
operator has fundamental flaws. Aviation work on
automation not only focuses on leveraging human
expect to encounter
performance and approaches to minimize errors.
It also emphasizes the need to keep operator skills more situations when
they must consider,
up-to-date, partly by education, partly by making
sure pilots are still able to fly manually, and partly
through strategically developed human centered
automation. or have embraced and
We think it’s no different for automotive techno-
logies.
As human drivers will continue to play active
trusted a priori, a mix
roles in driving their vehicles while managing an
increasing array of new or newly configured techno-
between human and
logies at their disposal, they can expect to encounter
more situations when they must consider, or have
automated control.”
embraced and trusted a priori, a mix between
human and automated control. These questions
become tangibly real for drivers of cars equipped
with advanced systems.
Millions of drivers over the next 10 years will not
only have to ask what their vehicle is able to handle,
but be prepared and comfortable answering them
with literal life-or-death certainty. We at Veoneer
see not just the opportunity but the necessity of
helping them.
Democratizing safety technology so that it
benefits the greatest number of people as soon as
possible is a new way of looking at our journey to
full automotive autonomy. We believe that such a
development can be enabled by a scalable safety
approach that puts each new safety innovation
wherever it can work effectively.
In pursuing this vision, we will build a context of
trust on our journey to 2030 and beyond.
14
You can also read
