Smart Industry 1/2018
Smart Industry 1/2018 - The IoT Business Magazine - powered by Avnet Silica
Smart Industry 1/2018 - The IoT Business Magazine - powered by Avnet Silica
You also want an ePaper? Increase the reach of your titles
YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.
<strong>Smart</strong> Business Title Story: Self-driving cars<br />
The company has invested heavily in<br />
research involving machine learning,<br />
which Huang says is the “bottom-up<br />
approach to artificial intelligence” –<br />
and probably the most promising<br />
technology today. Machine learning<br />
requires the processing of huge<br />
amounts of data, and as it turns out,<br />
the company’s computative graphics<br />
processing units (GPUs) can do the<br />
job both faster and using less energy<br />
than the traditional central processing<br />
units (CPUs) found at the heart of<br />
most mainframe, desktop, and laptop<br />
computers today.<br />
The computative power of GPUs has<br />
increased as computer images have<br />
become more complex and, in 2007,<br />
Nvidia pioneered a new generation<br />
of GPU/CPU chips that now power<br />
many energy-efficient data centers in<br />
government laboratories, universities,<br />
and enterprises.<br />
It was almost by accident that the<br />
company became a big player in the<br />
nascent autonomous car sector, but it<br />
now plans to release the Nvidia Drive<br />
PX2 platform next year, describing it<br />
as the first "AI brain" capable of full<br />
autonomy.<br />
Nvidia’s approach is so revolutionary<br />
that other chipmakers are scrambling<br />
to catch up. Intel and AMD, two of the<br />
largest manufacturers of computer<br />
chips, have teamed up to pool their<br />
resources in order to head off Nvidia<br />
by developing a GPU/CPU combo of<br />
their own. In addition, Intel made the<br />
How a self-driving car works<br />
Signals from GPS (global<br />
positioning system) satellites<br />
are combined with readings<br />
from tachometers, altimeters<br />
and gyroscopes to provide more<br />
accurate positioning than is<br />
possible with GPS alone.<br />
Radar<br />
sensors<br />
Ultrasonic sensors may be used to<br />
measure the position of objects very<br />
close to the vehicle, such as curbs<br />
and other vehicles when parking.<br />
The information from all of the<br />
sensors is analysed by a central<br />
computer that manipulates<br />
the steering, accelerator and<br />
brakes. Its software must understand<br />
the rules of the road,<br />
both formal and informal.<br />
biggest acquisition of its lifetime<br />
early in 2017 when it paid $15.3bn for<br />
Mobileye, an Israeli-based specialist<br />
in sensors, mapping technology, and<br />
camera-based devices for advanced<br />
driver-assistance systems (ADAS).<br />
This is just one example of many that<br />
hint at the impending disruption<br />
autonomous vehicles will bring to the<br />
automotive and related industries.<br />
As for when self-driving cars will hit<br />
the mainstream, opinions still vary.<br />
But Tesla, Ford, Audi, General Motors,<br />
and Nissan are among those that<br />
believe cars operating without humans<br />
will be on the road within the next<br />
five years.<br />
This visionary or perhaps doom-laden<br />
scenario, depending on your view,<br />
is driving the whole supply chain to<br />
frantically figure out where to position<br />
themselves. Delphi, a UK supplier of<br />
proprietary automobile components<br />
and integrated systems and modules,<br />
acquired NuTonomy, a developer of<br />
autonomous driving (AD) software<br />
solutions for $450m in November<br />
2017. Delphi, which will soon change<br />
its name to Aptiv, has announced it<br />
intends to build a self-driving system<br />
it can sell to all the big auto makers.<br />
ADAS requires immense computing<br />
resources to provide higher levels of<br />
predictability and autonomy.<br />
Typical components are: radar, lidar<br />
(similar to radar but using laser<br />
light), camera, ultrasonic, vehicle-toeverything<br />
(V2X) wireless sensors<br />
Lidar (light dtecting and ranging) sensors<br />
bounce pulses of light off the surroundings.<br />
These are analysed to identify lane markings<br />
and the edges of roads.<br />
Video cameras detect traffic lights, read<br />
road signs, keep track of the position of other<br />
vehicles and look out for pedestrians and<br />
obstacles on the road.<br />
Radar sensors monitor the position of other<br />
vehicles neraby. Such sensors are already<br />
used in adaptive cruise-control systems.<br />
Ranking Autonomy Levels<br />
HUMAN DRIVER<br />
MONITORS DRIVING ENVIRONMENT<br />
0 1 2<br />
No Automation Driver<br />
Assistance<br />
3<br />
Partial Conditional<br />
Automation Automation<br />
The Society of Automotive Engineers<br />
(SAE) definitions of vehicle automation<br />
AUTOMATED DRIVING SYSTEM<br />
MONITORS DRIVING ENVIRONMENT<br />
4High<br />
Automation<br />
5<br />
Full<br />
Automation<br />
■ How autonomous is autonomous?<br />
The Society of Automotive Engineers (SAE) International<br />
Standard J3016 offers some guidance on this important<br />
question with a six-level schema to describe the range of<br />
scenarios that are possible between traditional, unaided,<br />
and fully automated driving.<br />
Level 0 No Automation. The full-time performance of all<br />
aspects of the dynamic driving task, even when enhanced by<br />
warning or intervention systems, is left to the human driver.<br />
Level 1 Drive Assistance. The driving mode-specific<br />
execution of either steering or acceleration/deceleration is<br />
conducted by a driver assistance system using information<br />
about the driving environment, with the expectation that<br />
a human driver will perform all remaining aspects of the<br />
dynamic driving task.<br />
Level 2 Partial Automation. The driving mode-specific<br />
execution by one or more driver assistance systems of<br />
both steering and acceleration/deceleration is conducted<br />
using information about the driving environment, with the<br />
expectation that a human driver performs all remaining<br />
aspects of the dynamic driving task.<br />
Level 3 Conditional Automation. The driving mode-specific<br />
performance by an automated driving system of all aspects<br />
of the dynamic driving task is conducted with the expectation<br />
that a human driver will respond appropriately to a request<br />
to intervene.<br />
Level 4 High Automation. The driving mode-specific<br />
performance by an automated driving system of all aspects<br />
of the dynamic driving task continues even if a human driver<br />
does not respond appropriately to a request to intervene.<br />
Level 5 Full Automation. The full-time performance by an<br />
automated driving system of all aspects of the dynamic driving<br />
task occurs under all roadway and environmental conditions<br />
that could have been managed by a human driver.<br />
photo © SAE International and J3016<br />
14