LIV - Learning Intelligent Vehicle | White paper 2019 - Communication and collaboration between driver and AI control in "moments of truth" will ...
←
→
Page content transcription
If your browser does not render page correctly, please read the page content below
LIV – Learning Intelligent Vehicle | White paper 2019 Communication and collaboration between driver and AI control in “moments of truth” will determine consumer adoption of new technologies.
Collaboration as the
Summary key to safety
With new automated driving technologies slated • Drivers (and passengers) must trust that As a vast percentage of the world’s 1.4 million which may take a long time to recover from, as
to become available to consumers in 2019 — most automated systems will make the right decisions vehicular deaths each year are attributed to human some drivers may not use the automation feature
notably enabling vehicles to function with limited • Said systems must have the capacity to discern error, it’s taken for granted that removing humans for a long period (thereby negating the safety bene-
human involvement in a greater number of speci- and respond to different driver skill levels and from the driving equation will reduce that number; fit of the feature). From a Veoneer Research point
fic conditions — the need will emerge to be able emotions, and since people can’t be trusted, the innovation man- of view, the way to help mitigate this risk is through
to not just use but trust automated assistance in date should be to automate driving fully, and the true collaboration. This does not only involve a give
• This human machine interaction must not
quickly emerging and complex situations. These problem will be solved. and take of vehicle control, but also shared con-
only sometimes perform within seconds, but
situations, which we call “moments of truth,” will We see two problems with this assumption: trol which allows automation and humans to work
the give-and-take must result in learning, with
determine drivers’ understanding of, and trust in, First, many of the errors made by human drivers together.
collaboration getting better over time.
automation. are errors of judgement, situation assessment, or
According to our ongoing research, the solution recognition, and not outcomes of their humanity.
will be found in robust and nuanced collaboration This challenge is particularly relevant to Put another way, automated functions face the
between drivers, occupants, vehicles, and the realizing the core safety benefit promised What are moments of truth?
same challenges; in fact, only a few of such driving
external infrastructure, in three key areas: by the automation revolution. errors in more complex driving situations are due • High-risk
to performance or non-performance errors that • High-emotion
automated vehicles are less likely to commit (Adrian
Lund, 2018).
Secondly, examined from another perspective,
human drivers are generally quite accomplished
drivers, and experience tends to make us even bet- Changing tech will
ter. That’s not to say that there isn’t vast opportuni-
ty and need for improvement in traffic safety — even
change drivers, too
one traffic death is too many — but it will be some For the foreseeable future, the experience of
time before we can achieve complete traffic safety driving will remain a dynamic system that presents
entirely devoid of human drivers; in the meantime, an almost infinite combination of externalities, like
human drivers will continue to play active roles in weather, road conditions, and connectivity, thereby
driving their vehicles while managing the newest rendering automation unable to be perfectly reliab-
technologies at their disposal. le in every circumstance. Simply put, drivers will still
That means consumers can expect to encoun- need to be able to drive, when necessary, making
ter more situations when they must consider, or driver understanding of the system within which
have embraced and trusted a priori, a mix between she or he is embedded, and the system’s understan-
human and automated control. These questions — ding of the driver, essential for achieving safety in
often asked almost philosophically, like pondering those circumstances that matter most, such as:
the time when AI will be able to recognize, say, a
stop sign in every possible environmental condition • “Corner cases” that go beyond a vehicle’s
— will become tangibly real for drivers of cars equip- normal operating parameters
ped with advanced systems, who must not just ask • Emergency alerts
them, but be prepared and comfortable answering
• Notifications and vehicle responses
them with literal life-or-death certainty.
Consumer adoption of autonomous technology • Engagement with real-time traffic assistance
will depend in large part on how these ”moments of • Presentation of in-vehicle telematics
truth” are experienced. If handled badly by automa-
tion, they’ll result in a large decline in driver trust,
2 3
Veoneer Research sees the leveraging of the skills the time (with over-the-air updates, for instance),
and experiences human drivers have gathered, and it’s not enough: a car needs to teach its drivers and
its integration with the added autonomous capabi- passengers what it is newly capable of doing, not
lities coming to market, as an integral component simply provide a notification pop-up that “these
of consumer adoption as well as safety of those changes have been made.”
technologies. We need true collaboration between The scope of the learning challenge is also
vehicle and driver that evolves over time. broader than what’s contained inside the vehicle,
The aviation industry’s experience with adop- and must extend to understanding of changes in
ting automation suggests the path forward. Not the ground traffic system as a whole. New tools for
only does work on automation focus on leveraging communicating need to be designed to connect
human performance and skills, but also on keeping drivers with other vehicles (V2V), and vehicles and
those pilot skills up to date; training is mandatory infrastructure (V2X). It’s not only about the human-
for pilots throughout their careers. Pilot training machine interaction inside the vehicle.
is also focused on handling automation and break-
downs in automation, and is adapted as systems
change, so pilots learn to handle various system The implications of
states. As in aviation, we need to incorporate some
measure of continuous learning in road vehicle
riding vs. driving
Trust means knowing that difficult
automation driving. Right now, drivers receive a Recent research indicates that when drivers are fa-
license and, thereafter, even as laws and vehicle ced with descriptions of various types of driving au-
capabilities change, there is no additional focu-
sed or opportunistic learning to address them.
tomation, they appear to only be able to distinguish
between riding and driving (MIT AgeLab white paper
situations will be handled successfully.
Though cars themselves are “taught” new skills all “Consumer Facing Automation Types Taxonomy,”
by Seppelt, Reimer and others, 2018). That is, either potential errors committed by human or automated
they understand themselves as responsible for drivers. There are no more fully manual vehicles,
driving the car and respond accordingly, or they see and while there is a mix of human and automated
that it is the car’s role to drive, so they can spend control that drivers may misunderstand, we need to
their time engaged in non-driving related tasks. find a way to handle them safely without causing a
Consumers do not see automation as consisting of breakdown in the driver-vehicle relationship.
several “levels”, but understand vehicle competence
as an indication of either the car or them having the Put another way, there must be trust,
responsibility for object detection and response which means vehicles must pass the tests
tasks. This, of course, is a risk to consumers using presented by moments of truth.
vehicles with automation. In such moments, where
risk may be high, the automation may not perform
as people expect, as they are not designed to be
purely riding or purely driving automation; this, in Examples of moments of truth:
turn, can cause drivers to stop using automation • Using automation in an unfamiliar
altogether, as they do not understand their role in environment
this relationship.
• First time uses of automation
For taxonomies related to consumer information
a distinction between riding and driving can be of • Driving while distracted
use, but for vehicle automation design the implica- • Changes in vehicle functionality
tions of such results should go even further to define
a perimeter of protection, or safeguard, against
4 5
Collaboration and This means the vehicle system needs to be capable
to cope with all events that might affect driver and
differing capabilities occupant safety, as well as driver and occupant
trust in the system. Not only does the driver need to
Another theme that will factor into adoption of trust automation to actually use it, but the automa-
driving technology this year going forward has to tion needs to trust the driver to handle situations
do with monitoring the vehicle automation when and communicate if it needs to interfere in normal
automation can perform many, but not all, driving situations. If not, drivers may feel annoyed. As we’ve
tasks (again, vs. riding in a fully-automated, discussed, these circumstances can be complex,
autonomous, vehicle). and emerge quickly, and a vehicle’s actions and
Currently, driver engagement in automated responses in such moments of truth will have a
driving systems amounts to little more than the large impact on how the driver will trust the vehicle
ability to switch the system on and off, and someti- in the future.
mes set parameters such as cruise control speed,
or initiating lane changes. In response, the system Moments of truth are when drivers’
shows an icon if it is active, sees a lead vehicle, or understanding of, and trust in automation
recognizes lane markings. This communication affects the way(s) they decide to use the
between driver and system is basically a visual or system.
haptic monologue during a short time window, not
a dialogue within a shared context of collaboration When driving, this means the vehicle system will
and experience. need to keep the driver engaged in the driving task
by collaborating. That way, the driver will not be
made to suddenly have automation interfere in a
potentially surprising situation, but will have learned
during normal driving that they and the vehicle need
The future of trust to cooperate. Such driver-vehicle collaboration can
constitute establishing a safeguard and perimeter
of protection where automation, rather than human,
does the monitoring. Automation, after all, can mo-
Lack of trust is already a major roadblock for control and lane keep alerts. Better communication nitor tirelessly for extended periods of time.
adoption for currently available vehicle tech. Many about capabilities and understanding of driver and Consider a situation in which a child in the back
drivers disable vehicle automation, such as lane vehicle roles are tools; trust is the outcome. seat drops a toy and erupts in a fit, drawing the
keep assistance or adaptive cruise control, citing Trust means continuing to use a system or service driver’s attention away from the road, or a passenger
their belief that the functions are unreliable, even though it may fail sometimes, as failure is a begins to feel sick because of the precision by which
provide feedback at the wrong times, or are simply learning opportunity for intelligent systems much the vehicle can automatically speed up or slow down.
annoying. as it is for an intelligent creature. Such safeguarding will need to affect vehicle
There’s no reason to believe that the introduc- In the future, automated vehicles will adjust to dynamics by reducing risk and making sure that if
tion of more sensors or services alone will automa- driver preferences and learn as they drive, rather the driver’s attention is directed to a non-driving
tically address driver understanding and expecta- than only providing the binary option of turning on From the consumer and user perspective, this related task, the vehicle can still keep its occupants
tions. The ultimate remit for autonomous driving or off. Vehicles will even take initiative as an active makes it very difficult to communicate system and vulnerable road users safe by reducing speed
technology is to innovate trust. system rather than passively awaiting commands. capabilities. Drivers use trial-and-error for figuring and driving more conservatively. This collaboration
This means innovating not only when but how Over the next few years, understanding the human out what the system is able to do, then make their between driver and vehicle will be a partnership,
notifications are shared, but going beyond war- machine interaction in moments of truth will help assumptions of whether they can perform non-dri- based on trust, that both parties will have a hand in
nings to encompass the design of system actions us develop and deliver those systems and tools. ving tasks safely. Instead of reducing risk, such building. And, even though a vehicle might perform
as well as system availability. If the automated and behaviors increase it for a period of time as drivers within operational tolerances, the movement could
human actors in a vehicle can make reasonable Trust means acknowledging and sharing figure out how and if they are “driving” or “riding”. well cause car sickness for one or more passengers,
assumptions of the others’ performance and skill, risk. Trust is a relationship. During the journey of vehicle automation develop- so it means adjusting vehicle dynamics and pos-
collaboration could be implemented using existing ment, the perimeter of protection we noted earlier sibly driving related visual cues to a point where the
functions in novel ways, such as adaptive cruise will be key. occupants feel well.
6 7
The vehicle needs to get to know not only its driver,
but also its occupants to handle moments of truth
The learning driver
successfully. The resulting two-way trust is the & learning vehicle
building block of collaboration, and makes good on
the safety promise of any system. People drive differently in different situations,
For adoption of advanced vehicle automation, whether determined by internal characteristics
altering the user experience dynamic to one of (tired, agitated, distracted), or external circumstan-
more collaboration will require significant innova- ces (some people are just uncomfortable driving in
tion in system functionality, communication, and rain or snow). It is vital for future human vehicles
system self-monitoring capabilities. The design of and systems to learn about individual drivers and
system logic and interfaces inside and outside the specific tolerances, if only to avoid false positives
vehicle will be as important as those functional and alerts that might cause them to disengage
improvements, too, since the human-machine safety or comfort systems.
” Vehicles will need to be more flexible in their
functionality depending on external situations,
much as humans are.”
interface (including the driving task) needs to provide This does not mean that we will need to sacrifice
a real-time understanding of what the other can understandability or predictability, but rather
and cannot do (and perceive); otherwise, a moment explore ways that drivers can instruct their cars
of truth situation will cause a breakdown in two-way on what to do in a more nuanced fashion that suits
trust between drivers and their vehicles, and make their personalities and proclivities. This also goes
subsequent uses of said systems less likely. for the vehicles themselves, so learning how to be
clear about things as simple as software or map
Collaboration requires intelligence and updates should key into driver expectations and
learning, from the driver as well as the preparation.
vehicle. The outcome is user adoption –
drivers actually using the systems in their Vehicles will need to be more flexible in
vehicles. their functionality depending on external
situations, much as humans are.
Further, drivers will need to learn how to trust their
vehicles as well as understand their own roles.
Successfully addressing these factors could well
be the key to determining consumer adoption of
autonomous driving tools, not to mention delivering
on the safety benefit promised from their use.
8 9
Meet LIV3.0
LIV3.0 is a research platform intended to enable the For instance, LIV can recognize when a driver ap-
study and design of such driver-vehicle collabora- pears preoccupied or distracted, and automatically
tion, understanding, and trust. increase the vehicle’s following distance – and if the
LIV — the Learning Intelligent Vehicle — is an driver appears confused, can explain its decision,
artificial intelligence equipped car that can under- helping the driver and passengers learn about tech
stand and respond to context, using external and functionality. When traveling with LIV, drivers won’t
internal sensing combined with complex algorith- get annoyed by too frequent warnings and shut sys-
mic AI to create a unified assessment of what is tems off, but will only perceive the systems when
going on with the occupants, vehicle, and driving relevant, increasing both functional and perceived
situation, and then acting based on this evaluation. safety benefits.
At LIV’s core are deep learning algorithms that LIV’s learning and interaction with human beings
enable effective communication, including sensing inside as well as outside the vehicle will impact
driver gaze, emotion, cognitive load, drowsiness, state-of-the-art of safety development more bro-
hand position, and posture, and then using this adly, ranging from automated emergency braking,
information with data on the external environment to precautionary advanced driver assist systems
to yield driving experiences that are not only safer, (ADAS) and self-driving vehicles.
but feel that way, too. In 2019, we expect to learn more about the
requirements on our systems from a user perspective,
LIV represents the vision of Veoneer being able to communicate the information humans
Research, and will inform the premium need for true collaboration with an automated vehicle.
level of our future safety products. We will also learn about how this collaboration
should materialize, and what effects it will have on
the driving task.
We look forward to you joining us on
this journey!
10 11
veoneer.com
You can also read
