版权说明:本文档由用户提供并上传,收益归属内容提供方,若内容存在侵权,请进行举报或认领
文档简介
Customer-
CentricInnovationin
Automotive12Agenda>
Porsche
Engineering
Company
Overview
>
Engineering
Ecosystem
&
Development
Loop>Use
Cases•SignalFoundationModel•AI-CorrectedRace
GPS•Decoding
theFleet(AgenticDiagnosis)>Summary•
Function
andsoftware
development•
Software
integration
and
software
quality•
BigData
and
Artificial
Intelligence•Testing
and
simulation•System
development
and
integration•Testing
and
simulation•
Function
and
software
development•
Big
Dataand
Artificial
IntelligenceGermanyWeissach,
Bietigheim-Bissingen,
Leipzig,
Mönsheim,WolfsburgPorscheEngineering
US
is
integrated
into
an
international
networkof
competence
centers
focusing
on
digital
technologies.•
Headquarters
oftheInternational
Group
at
thePorsche
R&D
Center
Weissach•Vehicle
technologies
and
integration•
Function
and
softwaredevelopment
incl.software
architectures•Big
Data
and
Artificial
Intelligence•
Testing
andsimulationCzechRepublicPrag,
Ostrava•Vehicle
testing
and
development•
Driver
assistance
systemsand
highly
automated
drivingfunctions•V2X
testing•
E-mobility•
Endurance
testing
and
quality•
Driving
dynamics•
Connectionof
virtual
and
real
testingRomaniaCluj,
TimișoaraItalyNardò
Technical
Center•
China-specific
solutions•
Function
and
software
development•
Connectivity•
Quality•Testing
andsimulationChinaShanghai,
PekingUSALos
Angeles•
Market-specific
vehicle
functions•
Digital
services•
Connectivity
andinfotainment3FOCUSHERITAGELOCATIONSEMPLOYEES100%
subsidiary
of
Dr.Ing.h.c.F.Porsche
AGFoundedin1931byFerdinandPorsche
inStuttgartPEUS
as
Part
of
Porsche
Engineering
Group:
Company
OverviewinGermany,
Italy,
CzechRepublic,Romania,China
andUSA.Technologies
for
theintelligent
and
connectedvehicles
of
the
future1.700EmployeesPEG
ManagementDigitaltechnologiesMarkus-Christian
Eberl(CEO)VehicleexpertiseDirkPhilipp(CFO/COO)DirkLappe(CTO)13Locations4High-voltage
system
development
and
e-mobilityEnd-2-end
electronics
architectureChassis
system
developmentFunction
development
and
testingComplete
vehicle
development
Cayenne
CoupéAI
in
engineeringOur
Experience
in
theField
of
Vehicle
and
System
DevelopmentKnows
Many
Examples.5Porsche
Engineering
x
Sebastian
SteudtnerCommercial
vehicle
developmentX-BOW
with
dual-clutch
transmissionHigh-voltage
system
development
919
HybridChassisDemonstratorBattery
system
development
SeabobDesign
and
styling
of
Material
HandlingPorsche
Engineering
Group
(PEG)Projects6No
Matter
What
We
Develop,
We
Always
Keep
anEye
on
theIntegration
Throughout
The
Complete
Vehicle.
Quality
Included.Vehiclebody
systems
and
functionsE-mobility
and
high-
voltage
systemsE/E
architecture
ASPICE|
ISO26262|
ISO9001|
TISAXIntelligenceSimulation
and
testingChassis
systems
and
functionsComplete
vehicleintegrationHighly
automated
drivingDrive
systems
and
functionsDevelopmentConceptSeriesConnectivityArtificial7PROBLEM
DESCRIP
TIONController-centric
workflows→Longiteration
cyclesOffline
recordings→Hand-operated
dataoffload
(USB/loggers)Manual
measurement
setup
&
configuration→High
effort,
lowrepeatabilityFleet
testing
at
scale→Hard
tomanage,hard
to
compare
resultsResearch
ideas
stayed
long
in
lab
environments→No
finalConfidencebecause
of
missing
field
testsTransforming
from
lab
to
realworld
evaluation
in
the
vehicle
throughour
Cloud
Connected
Realtime
EcosystemWe
smartly
use
our
ecosystem
to
reduce
the
‘Time
To
Car'△8AI-enabledEdgeDeviceStandard
Vehicle
Architecture01
02
03
ANALYZEDeployOverview
of
Our
DevelopmentLoop EXECUTE&COLLECTEVALUATEOPTIMIZE
Cloud
Processing
/
Data
Center
DATA
LAKE
>
NEURAL
NETUpload9Edge
DevicesCloud
ResourcesOn-premise
ResourcesCentralizedDataLakeExperiment
Tracking&ModelRegistryScalableProcessingPlatformIoTEdgePlatform
withFleetMonitoringTechnical
Components
in
theDevelopmentLoopTest
Vehicleequipped
withadditional
EdgeDeviceCapabilitiesCloud-connected
VehicleMiddlewareNVIDIADGX
SystemswithShardingCapabilitiesWorkloadManagementTRAINEDMODELS&
DATASensor
Data&
Model
ResultsAzureIoTHubAzureMODELS10Start
at
the
Vehicle:
the
Edge
Device
as
the
Data
SourceDevices12CustomHardwarePlatformpoweredbyNVIDIA
JetsonHardware
The
CarDataBox
Serves
as
Edge
Device
in
OurAI
Ecosystem.ApplicationLayerConnected
VehicleMiddlewareSoftware13CoreRuntimeCANInterpreterFlexRayInterpreterAutomotiveEthernetCameraInterpreterConfigurationManagementApplicationManager
Custom
HardwarePlatform(NVIDIA
Jetson)Technical
Structure
of
the
CarDataBoxCore
SoftwareLayerApplicationLayerDockerConnected
Vehicle
MiddlewareHardware
LayerROSinterfaceROSinterfaceROSinterfaceMachineLearningComponentAzureUploader
ComponentReal-Time
ApplicationCore.
.
.
.
.14Remote
Application
Deployment
on
Edge
DevicesEngineerContainerRegistry
DEVELOP
&
DEPLOYEXECUTE
&
COLLECTANALYZE
&
OPTIMIZEAnalytics-,ML-
and
Processing-PipelinesRemoteDeploymentWeight
UpdatesDEVICEEvaluation
AppADAS
App▲ML
AppDataLakeAzureEDGE15Cloud
ResourcesOn-premise
ResourcesEdge
DevicesTest
Vehicleequipped
withadditional
EdgeDeviceCapabilitiesCloud-connected
VehicleMiddlewareCentralizedDataLakeExperiment
Tracking&ModelRegistryScalableProcessingPlatformIoTEdgePlatform
withFleetMonitoringTechnical
Components
in
theDevelopmentLoopNVIDIADGX
SystemswithShardingCapabilitiesWorkloadManagementTRAINEDMODELS&
DATASensor
Data&
Model
ResultsMODELS16Technical
Componentsin
theDevelopment
LoopAzureIoT
HubAzureDockerResources18Kubernetes
SpacesProject
space1Project
space
2Project
space
3Data
LakeIndexed
by•
Vehicle
signals•
Vehicle
derivate•
Locations•
Weather
•
…FleetAnalyticsSmallModel
Training
andModel
InferenceProcessing
the
CollectedData
in
the
CloudAzure
IoT
HubAzureNVIDIA
B200
on-premise>LargeModel
TrainingCurated
Datasets19Cloud
ResourcesOn-premise
ResourcesEdge
DevicesTest
Vehicleequipped
withadditional
EdgeDeviceCapabilitiesCloud-connected
VehicleMiddlewareCentralizedDataLakeExperiment
Tracking&ModelRegistryScalableProcessingPlatformIoTEdgePlatform
withFleetMonitoringTechnical
Components
in
theDevelopmentLoopNVIDIADGX
SystemswithShardingCapabilitiesWorkloadManagementTRAINEDMODELS&
DATASensor
Data&
Model
ResultsMODELS20Technical
componentsin
the
development
loopon-premiseResources22LocalInferenceContainerizationPartitioningof
GPUs
tosuit
workloadsDocker
/NVIDIA
ContainerRuntimeMiGPartitioningModel
deployment,
experiment
trackingLLMEvaluation/
MonitoringNVIDIA
DGX
PlatformvLLM
/
OllamaLLMInferencePoCServices
Model
TrainingDeepLearning
Engineers
DGX
Cluster
(B200)AgentDevelopersWorkloadScheduling&Model
Training
NIMTools
/
ProductsSLURM+PyTorch23Anend-to-end,consistentecosystemspanning
data
collection,
real-time
validation,scalable
training,and
edge
deploymentVehiclearchitecture-independentsoftwaredevelopmentenabling
flexibleadaptation
acrossplatformsFleet-scaleoperation,evaluation,and
data
collectionDirectin-vehicle
evaluation&Updated
withinminutesUnifiedglobal
AImodellifecycle,experiment
tracking,
&
governanceEfficiency
GainsFrom
our
Cloud
&
AI
Ecosystem>>>>>Azure
IoT
HubAzureDockercop
ik24Improving
vehiclepositioningaccuracy
duringhigh
dynamicdrivingAI-Agent
driven
vehicledata
analysisUSE
CASE
#1Signal
Foundation
ModelAI-based
scenario,patternand
anomaly
finderApplying
our
Ecosystem
toReal-World
ChallengesUSE
CASE
#3Vehicle
Data
AnalysisUSE
CASE
#2Race
GPS25USECASE
#1SignalFoundation
ModelAI-based
scenario,pattern
and
anomaly
finder26SignalFoundation
Modelfor
ADASScenario
Search,
Clustering,
andDescriptionCHA
L
L
E
NG
E•Largeamounts
of
unlabeled
signalrecordings•Rare/safety-critical
events
arehard
to
find,
compare,
and
quantify
across
fleetsAP
P
ROACHClustering
existing
databaseof
signalrecordingsinto
searchable
database
of
scenariosI
N
P
UTDriving
scenarios
as
time
series
dataO
UTP
UTSearchablelatentspaceLatentspace27Applying
Our
EcosystemDEVELOP
&
DEPLOY
EXECUTE
&
COLLECT
ANALYZE
&
OPTIMIZE•
LaneDetections•
ObjectDetection•
Acceleration/Speed
•
…Weight
UpdatesEngineerAnalytics-,ML-
and
Processing-PipelinesSignals
RecordedAzureDataLake28Description•
Extract
theimportant
featuresof
thescenario•
Representitby
aset
of
vectors
called
“embeddings”•
Embeddingis
acollection
ofcoordinatesinamultidimensionalspace•
Similar
scenarios
are
residingin
the
same
neighborhoodin
this
“Latentspace”Use
Cases
κ•
Similaritysearchingusing
embeddings•
Generatesimilar
scenarios
using
embeddings•
Similaritysearchusing
text•
ForecastingclassificationEmbedding*
Description**
[label]
A
car
at
a
mediumdistancein
themiddlelane
…Self-Attention(GroupedMulti-Query
Attention)
withKVCacheαliSignalFoundation
ModelTime
Series
DataTime
Series
“Foundation
Model”Large
Language
Model*Embedding–
thecoordinatesof
asinglepointinthelatentspace|**Description–
thelabelof
thecurrentscenariorepresentedby
the
embeddingFeedForwardSwiGLURMSNorm Softmax
Linear
RMSNorm
Domaininput
signalsEncoder
DecoderEmbeddingsLatentspaceLatentSpaceDistribution
RMSNorm
νµσ2Sample29SignalFoundation
ModelInitial
application
to
ADASGenerate
Scenesgiven
an
encoded
scene,generate
signals
for
similar
situationsSearch
Scenesgiven
aprovided
scene,search
for
similar
ones
in
a
databaseDescribe
Scenegiven
anencoded
scenario,
describein
natural
language30SignalFoundation
ModelBenefitsClosed-loop
improvement
from
deployment
to
optimizationTurn
rawrecordingsinto
a
searchable
scenario
databaseFaster
root-cause
analysisthrough
similarity
searchGenerate
scenario
variants
to
expand
test
coverage31USECASE
#2
AI-Corrected
Race
GPSforhigh-dynamic
driving32Track
from
Standard
GPSTrack
from
HighPrecision
GPSGO
A
L•Predict
andcorrect
GPSposition
error
directly
from
vehicle
sensor
datausingmachine
learning•
Achieve
accuracy
close
tohigh-precisionDGPSCHA
L
L
E
NG
E•Road
car
GPSnoise:
5-10muncertainty•Navigation
systemsleveragemapmatching&plausibility
checksHigh
Dynamic
Driving
Reduces
GPS
Accuracy•ClassicalGPScorrections
failinhigh
dynamic
driving
situations33SolutionLSTM
toincreaseGPS
accuracy
based
on
series
signalsDEVELOP
&
DEPLOY
EXECUTE
&
COLLECT
ANALYZE
&
OPTIMIZEEngineerSeries
Signals
Recorded•
High-precisionGPS•
Steering
/
Yaw
Angle•
Acceleration
/Deceleration•
IMU(InertialMeasurementUnit)
EDGE
DEVICEAnalytics-,ML-
and
Processing-PipelinesAzureWeightUpdates▲DataLake34Description•
AIbased
SensorFusion
toimprove
accuracybasedon
dynamic
features•
Considers
timehistory
of
vehicle
trajectory•
Targetslowresourceconsumption
to
allowin
vehicleprocessingGPS
Accuracy
ModelLSTMModelGPS-specificlossfunctionAI-Corrected
GPSHighPrecisionGPSINPUT
OUTPUTStandard
GPS35Track
from
Standard
GPSTrack
from
HighPrecision
GPSAI-corrected
GPSPositionAccuracy
boost:Achieves
a~95%
error
reductionGPS5–10m
→
~1–2m
(~1.4
m
RMSE)
Real-world
proof:Validatedon
differentcars(e.g.,
Porsche
911)
→
track-qualityracingline&lap
timingBasedon
standard
vehiclebus
signals
high-precisionGPS
ground
truth(trainingonDGX,
edge-ready)Benefits:
Increased
GPS
Accuracy
Through
LSTMl·36USECASE
#3Decoding
the
FleetAI
Agent-Driven
VehicleData
Analysis37AP
P
ROACH•
Anomaly
detection
on
adiagnosisprotocol
level•LargeProtocol
Transformer-Decoder
ondevelopment
fleet
protocols•Highlevel
of
automation
withmodular
AI
workflowsAI-driven
DevelopmentFleet
Insights
Reduce
Time
to
ReactionCHA
L
L
E
NG
E•Largeamounts
andmanymodalities
of
vehicle
data•
Ambiguousrelationshipsbetween
diagnosisinformation•High
efforts
arounddomain
specific
analyses38Our
solutions
integrate
into
the
development
dataloopand
enable
early
diagnosis.Low-codeworkflowappCustomizableanalysisinsightsIssueresolution
/development
measuresMl-driven
exploration
&
detectionData
feedbackloopsTimeseries
anomaly
detectionDevelopment
fleet
vehiclesAgentic
AI
AnalysisDatalake39Multi
AgentExpert
Process
Provides
Granular
Diagnosis
InsightsADAS
agentChassisagentBattery
agent…Rootcausediagnosticinsights
/
温馨提示
- 1. 本站所有资源如无特殊说明,都需要本地电脑安装OFFICE2007和PDF阅读器。图纸软件为CAD,CAXA,PROE,UG,SolidWorks等.压缩文件请下载最新的WinRAR软件解压。
- 2. 本站的文档不包含任何第三方提供的附件图纸等,如果需要附件,请联系上传者。文件的所有权益归上传用户所有。
- 3. 本站RAR压缩包中若带图纸,网页内容里面会有图纸预览,若没有图纸预览就没有图纸。
- 4. 未经权益所有人同意不得将文件中的内容挪作商业或盈利用途。
- 5. 人人文库网仅提供信息存储空间,仅对用户上传内容的表现方式做保护处理,对用户上传分享的文档内容本身不做任何修改或编辑,并不能对任何下载内容负责。
- 6. 下载文件中如有侵权或不适当内容,请与我们联系,我们立即纠正。
- 7. 本站不保证下载资源的准确性、安全性和完整性, 同时也不承担用户因使用这些下载资源对自己和他人造成任何形式的伤害或损失。
最新文档
- 关于2026年新市场拓展策略的讨论信函(6篇)
- 电解质的电离 课件 -2026-2027学年高一上学期化学鲁科版必修第一册
- 煤矿三违界定标准培训课件
- 煤矿生产安全事故隐患排查治理制度建设指南培训
- 肿瘤患者濒死期护理临床实践指南总结2026
- 面向大学生的射艺传统射箭文化科普
- 语文一年级下册《姓氏歌》
- 2026届井研县四年级数学下学期期末监测试题含解析
- AI大模型GEO服务商有哪些?2026年靠谱厂商实力评测与选择指南
- 银行业专业人员中级职业资格考试(银行业法律法规与综合能力)模拟题库及答案(黑龙江省2026年)
- 订单专员奖惩制度及流程
- 《耳鼻喉科鼻部手术诊疗指南及操作规范(2025版)》
- 2025北京丰台区初一(下)期末语文试题及答案
- 放射性肺纤维化诊疗指南(2025年版)
- 行业国际技术转移案例
- pcr实验室规范制度及流程
- 2026年中国邮政速递物流管理面试问题集
- 齐柏林飞艇课件
- 医防融合视角下的慢病防控体系
- DB64∕T 2171-2025 粉煤灰路基填筑应用技术规范
- TCWEA19-2023水利水电工程生态护坡技术规范
评论
0/150
提交评论