SCALABLE, SAFE UND MULTI-OEM CAPABLE ARCHITECTURE FOR AUTONOMOUS DRIVING.
9th Vector Congress | Simon FürstStuttgart | 21-Nov-2018
WE BELIEVE ONLY A FEW PLATFORMS WILL SURVIVE THE RACE FOR AUTONOMOUS DRIVING. THIS IS JUST ONE REASON WHY WE AIM FOR A MULTI-OEM PLATFORM.
9th Vector Congress | Simon Fürst | Scalable Architecture for Autonomous Driving | 21-Nov-2018
PAST FUTURE
Within just a few years only two operating systems for smartphones got established in the market: Android & iOS. All others died out!
Just as in the smartphone operating system market, only a few Autonomous Driving Platforms will succeed.
BMW TACKLES AUTONOMOUS DRIVING CHALLENGES WITH STRONG PARTNERS.
9th Vector Congress | Simon Fürst | Scalable Architecture for Autonomous Driving | 21-Nov-2018
Leading Technology Partners
• More than 100 years of automotive design and production.• Premium ADAS customer experience.• Advanced vehicle electronics integration.
• More than 4.7 million vehicles per year.• Advanced vehicle electronics integration and redundancy
architecture capabilities.• Strong US and European footprint.
• #1 semiconductor manufacturer.• Broad support for OS and safety.
• Leading automotive computer vision technology provider.• Extensive AI expertise.
Leading Integration Partners
• Specialized know-how within Automotive industry.• Reliable system integration partners.• International footprint with high standard.• State of the Art technology providers (e.g. LiDAR)
Leading Automotive OEMs
Resources
Team Technology
Market Relevance
State of the Art
ConceptProduct
Focus
A SCALABLE, SAFE UND MULTI-OEM CAPABLE ARCHITECTURE NEEDS TO INTEGRATE THE FOLLOWING MAIN CHALLENGES OF AUTONOMOUS DRIVING.
9th Vector Congress | Simon Fürst | Scalable Architecture for Autonomous Driving | 21-Nov-2018
HAR
DWAR
E &
INTE
RFAC
ES Motion Control, Odometry
High Performance and safe automotive silicon.
SOFT
WAR
E Driving PolicyPrediction Environment Model
DATAData Center & Big Data Processing
Al training andAlgorithm Evaluation
Data Collection
360° Sensor perception, HD-Map, Localization
COMMON SCALABLE SENSOR CONCEPT MAXIMIZES VALIDATION SYNERGIES.
Short Range Radar Side
6 UltrasonicSensors
Highway Pilot Level 3
Urban Pilot Level 4/5
6 UltrasonicSensors
Surround View Cam Surround View Cam
SurroundView Cam
SurroundView Cam
Front-View CameraFront Radar
Short Range Radar Side
Short Range Radar Side
Short Range Radar Side
LIDAR Front
Far-Range Front-ViewCamera
Short RangeSide Radar
Short RangeSide Radar
Rear Radar
Rear Radar
Rear Driving Camera
Extended Front-ViewCamera for Urban
Driving CamSide Rear
Driving CamSide Front
Driving CamSide Rear
Driving CamSide Front
LIDAR Side
LIDAR Side
LIDAR Rear
LIDAR Rear
RadarLIDARCamera
ADASLevel2
9th Vector Congress | Simon Fürst | Scalable Architecture for Autonomous Driving | 21-Nov-2018
A MULTI OEM API CONNECTS THE COMMON SW-STACK WITH THE OEM SPECIFICSOFTWARE AND HARDWARE.
9th Vector Congress | Simon Fürst | Scalable Architecture for Autonomous Driving | 21-Nov-2018
Vehicle dynamics integration platform
nominal integration platform
FAD
HAD
ADAS cam front
side radars
front radar
Rear radars
Side radars
LidarSide/Rear
NavigationECU HD Map
Cams side
Lidar Front
OnboardOffboard
TelematicsECU
Test fleet Backend for Customer Cars incl. HD Map
Learning Map
Image-/Sensordata
Motion Control
EgoMotion
Cams parking
Cams Front+rear
Steering
Brake
Engine
RedundantSteering
RedundantBrake
Ultra sonic
driver cam
Interface and
toolchain
Middleware, Operating system
Reference design / performance µP
Middleware, Operating system
Reference design / performance µP
Com
pute
r Vis
ion
Sele
ctor
Grid +ML Fusion
HD MapRepresentation
Motion Control fail-operational
Trajectory Following Controller
2nd ASIL Channel
Primary ASIL Channel
Fail degradad integration platform
Development Backend incl. ML und Simulation
Fusion Driving Policy Validator
Fusion Driving Policy Validator
Fusion Driving Policy
L2 SW-stack is used asL3 fail degradion stack
Common SW-Stack OEM-specific SW
Traj
ecto
ry F
ollo
win
gCo
ntro
ller
BMWS SAFETY VISION FOR AUTONOMOUS DRIVING.
9th Vector Congress | Simon Fürst | Scalable Architecture for Autonomous Driving | 21-Nov-2018
Targeting for maximum safety and significant improvements compared to today‘s average human drivers. • Avoid accidents under any circumstances, no matter who is responsible for causing the accident.• If accident is unavoidable, minimize human severity.
To achieve this vision we are developing a scalable safety concept for Highway-Pilot (L3) and Urban Pilot (L4/5) applications, together with strong and safety-oriented OEMs, Tier1s and Technology Partners.
Effects of automation on the driver need to be taken into account
THE TEN SAFETY COMMANDMENTS FOR AUTONOMOUS DRIVING.
9th Vector Congress | Simon Fürst | Scalable Architecture for Autonomous Driving | 21-Nov-2018
Driver‘s responsibilities: Driver monitoring and clear mode awareness necessary
Since the system is responsible to fulfil the driving task, functional safety requires redundancy
If system limits are reached the system need to minimize those risks or performs a safe TOR with sufficient time
Behavior of the system in traffic needs to be predictable. All valid traffic rules need to be taken into account
Security: Protection against manipulation of the system is mandatory
Driver initiated activation and deactivation needs explicit driver intend
If the driver does not respond to a system initiated TOR a minimum risk maneuver has to be performed
Multi stage validation including simulation (e.g. Pegasus: 240 Mio km) is necessary to ensure that the safety goals are met
Data recording while the system is active is necessary to recognize and document unusual events
TOR: Take over request
FUNCTIONAL SAFETY FOCUS FOR HIGHWAY PILOT AND URBAN PILOT.
9th Vector Congress | Simon Fürst | Scalable Architecture for Autonomous Driving | 21-Nov-2018
ASIL D
ASIL B
Fusion of LIDARs, RADARs and cameras with a range up to 300m
Safe and high precision positioning
Safe Localization
HD-Map
Safe HMI for a clear mode awareness
Road condition previewIncluding hazard warning
Emergency stop assistantMinimum risk maneuver
Highway-Pilot and Urban Pilot details see next page
ASIL C
ASIL QM
Safety Features like Automatic Emergency Braking
Teleoperation of vehicle
Fail-degraded ECU
Nominal ECU
Selector + Trajectory Following Controller
Cameras
2nd ASIL channel
Fail-degraded Electrical Network
Main Electrical Network
SAFE Fail-degraded Channel
Validator
ASIL D
Trajectory Following Control + Motion Control
Motion Control
Vehicle dynamics ECU
ASIL BFUNCTIONAL SAFETY CONCEPT IS PREDICTIVE AND AGNOSTIC TO HARDWARE AND MAP. IT IS APDAPTABLE TO ANY SOFTWARESTACK.
Main Channel
Driving PolicyPerception
ValidatorRadars
Cameras
Radars
Lidars
LocalizationHD MAP
Driving PolicyPerception
Prediction
9th Vector Congress | Simon Fürst | Scalable Architecture for Autonomous Driving | 21-Nov-2018
Page 11Subject | Department | Date
THE CHALLENGEXPAD ECU FAMILY - FACTS AND FIGURES
4 Different SOC architectures Infineon AURIX, Intel Denverton, Intel Xeon, MobilEye EQ5 (MIPS)3 Internal software suppliers EV, EF, JC 5 External Software Suppliers: Intel / MobilEye Aptiv N.N. (Adaptive AUTOSAR) N.N. (Safe Linux) N.N. (Supplier for GNSS positioning engine)25 software images (25 diagnostic addresses) 2 mPAD 7 hPAD 16 uPAD
21 Adaptive AUTOSAR images!
Page 13Subject | Department | Date
AUTOSAR AT BMWSP2021 CONCEPT CHANGES IN BMW SYSTEM SOFTWARE
Seite 13
BMW system software components will be redesigned in order to support different API requirements They split into a generic part and a platform specific adapter
Application Mode BAC 4 SP2021 ABAC SP2021 GenIVI in MGU in 2022
GENIVI ADAPTERSTDDIAG
GENERIC PARTSTDDIAG
ADAPTIVE AUTOSAR ADAPTERSTDDIAG
CLASSIC AUTOSAR ADAPTERSTDDIAG
Exam
ple:
BM
W S
tdDi
ag
MAIN BENEFITS OF SAFETY CONCEPT FOR AUTONOMOUS DRIVING.
Benefits of collaboration on safety concept Underlying concept
Automotive grade safety concept and transparent implementation approach for partners and regulators. White box development and shared code basis.
Maximizing availability of AD feature in diverse traffic environments of different regions of the world.
Permanent analysis and validation of the planned driving actions.
Combining LIDAR, RADAR and camera to achieve maximum advantages by the following multi sensor fusion.
Functional safety for AD features Minimization of common cause failures through hardware and software diversity.
Increased flexibility to integrate and combine any ASIL B platform software Hardware, Software and Map agnostic approach.
9th Vector Congress | Simon Fürst | Scalable Architecture for Autonomous Driving | 21-Nov-2018
THANK YOU FOR YOUR ATTENTION. LOOKING FORWARD ON YOUR FEEDBACK AND QUESTIONS.
9th Vector Congress | Simon Fürst | Scalable Architecture for Autonomous Driving | 21-Nov-2018