Page 27 - ATZ WORLDWIDE

BILLIONS OF MILES OF ROAD AUTHOR
TESTING ARE NECESSARY

According to experts, autonomous vehi-
cles will change industry and society
itself as much as the nineteenth century
shift from horse-drawn carriages to auto-
mobiles did. Developing autonomous Sandeep Sovani, Ph. D.
vehicles is an exceptional challenge. It is Director Global Automotive
requires ambitious new developments in Industry by Ansys Inc. in Canons-
sensing technologies, machine learning burg, Pennsylvania (USA).
and Artificial Intelligence (AI), which are
not only unprecedented in the automo-
tive industry, but in all other industries
as well. The main task is to replace a
human driver with artificial intelligence.
An autonomous vehicle’s computer
essentially needs the ability to recognise simulations. Thousands tests can be
other vehicles, pedestrians, road signs, completed within the time and budget
road markings, trees, buildings, traffic required for a single physical test, thus
lights and a variety of other things that greatly accelerating technology
we encounter every day during a trip, development.
and also under poor driving conditions Virtual simulation provides three sig-
like in the dark, rain or snow. nificant benefits: First is faster time-to-
This problem is almost unsolvable market as simulation is conducted in a
with traditional rule-based computer virtual environment and therefore, the
algorithms. Instead, neural-networks time required is considerably shorter
and machine-learning methods need to than physical prototyping and testing.
be used. In these methods the computer Second is reduced cost since simulations
is trained rather than programmed. Driv- are considerably more economical than
ing is, however, such a complex task that physical tests. Furthermore product
an immense amount of training hours quality is enhanced due to deep insights
would be needed to make a computer into the underlying physics which deter-
drive as safely and reliably as an average mine design and operation of a product.
driving human. An investigation done by Simulation plays a key role in solving
the Rand Corporation [1] indicates that quality issues upfront. Considering these
© chombosan | iStock
an autonomous vehicle must complete advantages and faced with the “daunting
billions of miles of road tests to train its challenge” of billions of test miles, it is
AI to the same level of an average driver. not surprising that manufacturers of
In the autumn of 2016 Akio Toyoda, Advanced Driving Assistance Systems
president of the Toyota Motor Corpora- (ADAS) and autonomous vehicles
tion, mentioned that 8.8 billion of test consider simulations as an essential
miles, including simulation, would be technology with which they can achieve
needed to develop self-driving cars [2]. their development goals. They are focus-
On the other hand, Google recently ing on the following six aspects of simu-
completed a cumulative total of two lation, FIGURE 1.
million miles of testing with its fleet of
self-driving cars over the past six years DRIVING SCENARIO
[3]. At that rate, it will take millennia to SYSTEM SIMULATION
develop a usable autonomous car.
Comprehensive driving scenario simula-
tions can be carried out with a system
NEED AND BENEFITS OF
SIMULATION behavioural model of an autonomous or
semi-automated vehicle. Such a vehicle
Engineering simulation therefore plays model includes all sensors, control sys-
an essential role by expediting technol- tems, drive systems and vehicle dynam-
ogy development. The automotive ics, placed in a virtual driving environ-
industry has been using simulation ment consisting of roads, buildings,
since a few decades. Virtual tests can be pedestrians, road signs, etc. In this sim-
completed on virtual prototypes with ulated environment, thousands of driv-
ATZ worldwide 09|2017 25