Three Questions
on ADAS/AD
Scenario Testing
;hEt iW %:0 7GenErio 7iQYPEtorƦ#
AVL Scenario Simulator™ is a continuous integration and
testing platform for ADAS/AD software that provides
simulation models and tool interfaces to create a virtual world
for reproducible scenario-based tests. This allows you to
test aYtonomoYs veLicles in reTrodYciFle, nearcrasL traƾc
situations in a virtual world without putting people at risk.
,ow HoeW it PinO into other %:0 tooPW EnH other QErOet tooPW#
1odYlarity and ƽe\iFility are among tLe Figgest strengtLs oJ
Scenario Simulator. Here we can differentiate between process
integration and simulation coupling.
For process integration, a powerful Python API allows full
aYtomation oJ tLe tool configYration and simYlation e\ecYtion
for easy integration into any existing CI/CD pipeline. If you
are looking for a holistic solution, AVL SCENIUS™ covers the
entire worOƽow Jrom reUYirement engineering, to scenario
creation and management, to test case planning, execution,
and reporting. Thanks to native support for the ASAM
OpenSCENARIO® standard, all modules also work on a
standalone basis.
For simulation coupling, the included models for sensors
and vehicle dynamics can be replaced with own models or
even third-party tools based on a vast library of open and
customizable interfaces. The core value here is that users can
create virtYal TrototyTes tLat reTresent e\actly tLe model fidelity
reUYired Jor a sTecific test TYrTose. 8Le TerJormance Trofiling
functionality helps users to analyze resource usage of the
coupled models and tools to optimize the performance of the
overall simulation setup. This avoids unnecessary computations,
and even a saving oJ Ż can translate to tLoYsands oJ eYros
when scaling to millions of tests in production usage.
;hEt EFoYt TErEPPePi^Etion Ɓ Ere there PiQitEtionW Jor the YWer#
The main parallelization limitations are the available machines
and licenses. For licenses, we offer incredible volume discounts
tLat end YT in a ƽat rate oJJering tLat allows Ynlimited Tarallel
executions. Our job scheduler can distribute the simulations to
mYltiTle worOers. 8Lis can Fe a set oJ worOstations in tLe oƾce
or a Kubernetes cluster in a company-owned or public cloud. In
our scaling tests, we encountered the limitations of public cloud
providers when requesting more than 1000 CPUs without prior
notice.
It is also important to mention here that AVL Scenario
7imYlatorƦ comes witL lowfidelity simYlation models tLat
provide a high-performance testing environment for planning
and control algorithms. This setup enables 10-20 parallel
executions, even on a notebook, resulting in a throughput
of 50k-100k scenarios per day. Meaning that parallelization
works even on local machines and does not depend on a cloud
inJrastrYctYre.ɸ
Interview with
Simon Terres
Solution Manager
ADAS
HOME OF INNOVATION
AUTOMATED AND CONNECTED MOBILITY
50 | 51