Marek Jersak «Autonomous Drive – From Sensors to Motion».
- 1. Marek Jersak Senior Director, Autonomous Drive Autonomous Drive Empowering the Mobility Revolution
- 2. Reality… Picture of congested area
- 3. www.luxoft.com Some “non-autonomous” driving statistics Safety 1.2 mln people die on roads worldwide every year. This is about one death every 30 seconds Accessibility 10s of millions of people are too young, too old, not trained, or physically unable to drive Time On average we commute 10% of our free time. Consider adding 8 more years to your productive lives! Money Cars are only used 6% of time, wasting 94% of cost
- 5. www.luxoft.com PERSONAL DIGITAL LIFESTYLE AUTONOMY ELECTRIFICATION SHARING ECONOMY The Mobility Revolution www.luxoft.com 5
- 6. Key Industry Trends Automotive companies become Software Companies. Autonomous Driving & ECU consolidation drives the need for new software platforms and talent. Sharing economy drives digitalization of customer experience for car industry. Eco-systems need to be built to satisfy Personal Digital Lifestyles, no longer “one company does all”. 1. 2. 3. 4. www.luxoft.com 6
- 7. Flexible & efficient engagement model Best use of resources Easily scalable Balancing cost & other factors NORTH AMERICA WESTERN EUROPE CENTRAL AND EASTERN EUROPE ASIA AND PACIFIC GLOBAL DELIVERY PLATFORM 13,000+ Employees Worldwide 300+ Active Clients, 40% from Fortune 500 $900M Est revenue FY’2018 23% Compound Annual Growth Rate Public listed on the New York Stock Exchange Continents4 20 Countries 41 Cities 38 Delivery offices Luxoft at a Glance
- 8. 2,700+ Employees Worldwide 32+ Active Clients, 40% OEMs <10 % Attrition 45% Compound Annual Growth Rate 4 new locations in FY18 12 years Automotive Practice Geographies 3 3 Continents 17 Countries 16 Delivery offices • APAC • Americas • EMEA Luxoft Automotive at a Glance
- 10. Co-creating Smart Solutions The world has become far too complex and far too diverse for any one company to be able to meet all the demands of customers. Dr. I.P. Parks, LG President & CTO, Opening Key Note Speaker at IFA, Berlin, Aug. 31, 2018 Co-creating smart solutions in practice – together with an ecosystem of technology, product, and platform partners. Two cornerstones: next generation silicon and software platforms www.luxoft.com 10
- 12. Autonomous Drive From Sensors to Motion
- 13. 2050 ?2018 L3 Conditional AUTOMATION Vehicle takes control most of the time Driver is required to take over in emergency L4 High AUTOMATION Vehicle fully autonomous under certain conditions. Driver is not required during defined use case L2 PARTIAL AUTOMATION Vehicle integrates detection. Driver ready to take control L1 DRIVER ASSISTANCE Vehicle provides driver infos/warnings. Driver has informed control L0 NO AUTOMATION Driver has full control 20302020 MONITORING FALLBACK L5 FULL AUTOMATION Vehicle fully autonomous. Occupants do not need ability to drive Passenger Mind Off Eyes Off Hands Off …Series Cars SAE Levels auf Automated Driving www.luxoft.com 13
- 14. Safety-critical vehicle control … Central Driving Unit Car Actuators Camera, Lidar, Radar This can happen with a level 2 system, when the driver assumes it is level 3.
- 15. AD SW Development Aspects www.luxoft.com 15 Lane Keeping Lane Changing / Exiting Intersection Parking … Emergency Stopping 2. Finite State Automata of Functions1. Logical / Function Architecture 3. HW / SW Architecture Global Route Planning Sensing HD Maps / Cloud Data Perception Motion Planning Control & Feedback Driver Model Smart Road Infrastructure 4. Development Process • ASPICE • Functional Safety • Agile Scaling Frameworks, CI / CD • Real-Time
- 16. Progressing towards L4 Autonomous www.luxoft.com 16
- 17. AD System Architecture www.luxoft.com 17 Interior / Exterior camera camera LiDAR RADAR RADAR RADAR RADAR RADAR camera camera camera RADAR RADAR LiDAR LiDAR Redundant AD Computer UPA UPA UPA UPA UPA UPA UPA UPA camera LiDAR RADAR camera camera Cockpit Computer Redundant AD Computer Body/GW Computer
- 18. Sensor Data Fusion – why is it important? Camera + colors, shapes, object recognition o 3D, dynamics Lidar + disances, relative velocities o resolution Radar + range, field of view, poor weather - resolution
- 19. AD Software Platforms High-Performance Cores (x86 or Cortex-A) Hypervisor Applications (demonstrators) POSIX OS AR Adaptive Platform Applications (demonstrators) SeveralGuestOSpossible(Linux,QNX,etc.) Shallowvirtualization(Linuxcontainers)possible Memory, Peripherals, Interconnect, Accelerators, etc. Platform SW ARA::COM High SW interface Low SW interface MCAL drv/libsBSP SOC vendor make or buy* buy* + some make buy* SOC vend. + customization make High-Performance Cores (x86 or Cortex-A) High-Performance Cores (x86 or Cortex-A) Classic Cores (Cortex-M/R or proprietary) High-Performance Cores (x86 or Cortex-A) Classic Cores (Cortex-M/R or proprietary) * Buy can also include OSS (open - source software) SOC and board User Software OEM middle ware Embedded OS AR Classic (full or reduced) AR-CP-RTE
- 20. Multi-domain SW Architecture Continuous integration incl. test and measurements Timing Analysis www.luxoft.com 20 Luxoft Autostar Adaptive Contributions
- 21. Hardware Software Unit Testing Hardware Software Board Integration Testing HIL and SIL tests Communication tests System Integratio n Testing Performance tests Vehicle Integration Testing Supplier Identification Design Transfer PPAP Production User Manual Warranty Product LaunchComponent Technical Specification Compliance standards Interface system details Use case definition Product Definition Functional safety Compliance standards System safety requirement / Goals Risk Assessment Safety Definition System Scope and Boundary Development Interface Agreement System Safety Program Plan Supporting Processes Plan Technical Safety Concept Development Interface Agreement Requirement Gathering Test plan development ISO26262 design consideration HW & SW Design, worst case, safety analysis Product Design and Development Luxoft activities Customer activities Customer and Luxoft Luxoft Partnered Product Development Partnership
- 23. KEY AGILE SCALING FRAMEWORKS Scaled Agile Framework (SAFe) - Much more detailed than LeSS and Nexus - Quicker start due to pre-defined practices - Specific focus on funding, P&L, portfolio management - Introduces new dedicated roles, more focus on personal accountability - Less focus on shippable product increment - Logical way to move slowly towards Agile for large enterprise companies Large Scale Scrum (LeSS) - Little overhead over Scrum - More of a lean thinking instrument than specific toolkit for scaling - Requires senior Scrum Masters and Product Owners - Success depends on the experience and mindset of people driving implementation - Effective for “green field” product development teams having necessary resources to establish proper practices Nexus (SPS) - Most similar to plain Scrum - Has dedicated integration team, reducing seniority requirements for regular teams - The “youngest” approach, doesn’t have too many case studies yet
- 24. Autonomous Drive Challenges & Remedies
- 25. Steven E. Shladover, Sc.D. California PATH Program Institute of Transportation Studies University of California, Berkeley Addressing Autonomous Drive Challenges Complexity of Self-driving Functions: Road Conditions www.luxoft.com 25
- 26. Steven E. Shladover, Sc.D. California PATH Program Institute of Transportation Studies University of California, Berkeley Addressing Autonomous Drive Challenges Complexity of Self-driving Functions: Idiots www.luxoft.com 26
- 27. Car Companies will need to process massive amounts of data www.luxoft.com 27 OEM factory in 5 years: 100 racks 15,000 racks It is estmated that putting L4 Autonomy on the road will require hundreds of PETA-Byte of Data
- 28. Highly-Automated Data Labeling Raw data to be labeled (e.g. lidar point cloud) Data visualization for humans Supplemental raw data for humans (e.g. video) Label prediction (automation) Integrated annotator workplace Humans fix label predictions Human org. process to validate the output Results: human-verified labels for lidar data START END improve accuracy improve UX for efficiency Speed and quality Add computer validation Cheaper, better results Add fusion-based label prediction
- 29. Driving Fallback (L3 – L4) www.luxoft.com 29
- 30. Teleoperated Driving (L5) www.luxoft.com 30 vehiclecontrol center Remote access to vehicle control
- 32. Vehicle Security Module safety- critical features control signals video stream vehicle data TMIS control center hacker vehicle communication channel e.g. WIFI www.luxoft.com 32
- 34. www.luxoft.com 34 Automotive in Silicon Valley Continental Automotive Lyft Hyundai Ventures Honda Research Google Auto Tech GM Advanced Technology Ford Silcon Valley Lab Valeo North America Uber Auto Toyota InfoTech Tesla Uber / Otto Nvidia Auto Nissan Research Center NIO (NextEV) BMW Group Technology Office Delphi Labs Zoox Porsche Digital Tesla Factory Bosch Research OEM Visteon Tier 1 Disruptive Apple Car Audi / VW Electronics Panasonic R&D SAIC Mercedez-Benz R&DNauto Lucid Motors Samsung Waymo