Physics Based Sensor simulation -...
Transcript of Physics Based Sensor simulation -...
Physics Based Sensor simulationJordan Gorrochotegui - Product Manager Software and Services
Mike Phillips – Software Engineer
Realize innovation.Restricted © Siemens AG 2017
Siemens offers solutions across all automotive mega trends
Autonomous Vehicles Electrified Vehicles Mobility
Engineering the next product not just the best product for the future
Connected Vehicles
12
3
45
6
7
Smart fleets & multi-modal
transport
Autonomous driving &
driver assist systemsElectric vehicles &
supporting technology
Connecting car, driver and
infrastructure
Restricted © Siemens AG 2018
2018.MM.DDPage 3 Siemens PLM Software
Real world
Vehicle under development
Validation and Verification framework for AVs
MiL / SiL / Cluster
HiL / DiL / ViL
Proving ground / field test
V&V environments
Digital Twin
“World”
Digital Twin
“Vehicle”
Design adaptations
(HW/SW)
1M – 10M scenariosRequirements
Multiple variants
Certification -
Homologation
Simulation definition
Requirements &
system architectures
10 – 100 scenarios
1k – 10k scenarios
Restricted © Siemens AG 2018
2018.MM.DDPage 4 Siemens PLM Software
Real world
Vehicle under development
Validation and Verification framework for AVs
MiL / SiL / Cluster
HiL / DiL / ViL
Proving ground / field test
V&V environments
10 – 100 scenarios
1k – 10k scenarios
Design adaptations
(HW/SW)
1M – 10M scenariosRequirements
Multiple variants
Certification -
Homologation
Simulation definition
Requirements &
system architectures
Digital Twin
“World”
Digital Twin
“Vehicle”
Restricted © Siemens AG 2018
2018.MM.DDPage 5 Siemens PLM Software
Example #1: MiL / SiL / Cluster
• Design space exploration using large scenario databases
• Virtual development, verification and robustness testing
• Optimized automated vehicle designs
V&V Environments
Laboratories / proving grounds / public roads
10 – 100 scenarios
1k – 10k scenarios
HiL / DiL / ViL
MiL / SiL / ClusterRun massive amounts of Prescan scenarios for
Automated Vehicle development and optimization
Restricted © Siemens AG 2018
2018.MM.DDPage 6 Siemens PLM Software
V&V Environments
Laboratories / proving grounds / public roads
10 – 100 scenarios
1k – 10k scenarios
HiL / DiL / ViL
MiL / SiL / Cluster
Example #2: HiL testing of central AD processing unit
• Training and evaluating Deep Neural Networks (DNNs)
• Virtual validation of automated driving processing units
• Accelerated automated vehicle development
PreScan synthetic sensor data injection for virtual validation of central AD processing units
Automated Driving processing unit
Automated Driving system validation
Restricted © Siemens AG 2018
2018.MM.DDPage 7 Siemens PLM Software
Example #2: HiL testing of central AD processing unit
Free space detection on Nvidia Drive PX2
Restricted © Siemens AG 2018
2018.MM.DDPage 8 Siemens PLM Software
Example #2: HiL testing of central AD processing unit
Object detection on Mentor DRS360
Restricted © Siemens AG 2018
2018.MM.DDPage 9 Siemens PLM Software
V&V Environments
Laboratories / proving grounds / public roads
10 – 100 scenarios
1k – 10k scenarios
HiL / DiL / ViL
MiL / SiL / Cluster
PreScan – HIL application examples
Restricted © Siemens AG 2018
2018.MM.DDPage 10 Siemens PLM Software
V&V Environments
Laboratories / proving grounds / public roads
10 – 100 scenarios
1k – 10k scenarios
HiL / DiL / ViL
MiL / SiL / Cluster
Example #3: Automated Driving physical validation
• Physical verification & validation services
• Certification of automated and connected systems
• Design consultancy for “next-generation” AD test facilities
TASS International Services and Siemens Testing Solutions for physical validation of automated and
connected driving technology
Restricted © Siemens AG 2018
2018.MM.DDPage 11 Siemens PLM Software
Development of automated vehicles requires realistic sensor data
DNN TrainingSensor developmentAlgorithm development Validation
Restricted © Siemens AG 2018
2018.MM.DDPage 12 Siemens PLM Software
? Realistic data
+ Inexpensive
+ Fast to acquire
+ Perfect repeatability
+ Physical sensor not needed
+ Annotation is free
+ Can be closed loop
+ Real data
- Expensive
- Time consuming
- Not (easily) repeatable
- Requires a physical sensor
- Must be annotated
- Open loop (recorded)
Two sources of sensor data
Recorded Simulated
Restricted © Siemens AG 2018
2018.MM.DDPage 13 Siemens PLM Software
Camera
&
Ground Truth
Radar
&
V2X
Lidar
&
Point Cloud
Restricted © Siemens AG 2018
2018.MM.DDPage 14 Siemens PLM Software
We use simulation to compute real world effects on actual sensors
Output verified & validatedRaw signal is computedPhysical artefacts are
faithfully reproduced
• Distortions
• Multi-bounce
• Time effects
• Weather
• …
• Lidar full-waveform signal
• Radar channel response
• Raw camera images
• …
• Verify against real sensors
• Validate for specific use cases
Restricted © Siemens AG 2018
2018.MM.DDPage 15 Siemens PLM Software
Physical Device World
Simulation Engine
Beam cross
sectionMTF
Color filter
array
Three key components are needed
Restricted © Siemens AG 2018
2018.MM.DDPage 16 Siemens PLM Software
Physics Based Lidar Simulation
Restricted © Siemens AG 2018
2018.MM.DDPage 17 Siemens PLM SoftwareEnsuring digital continuity, multi-domain traceability and functional safety of autonomous systems
Our solutions support autonomous vehicle development needsAcross all engineering domains