1、World Leader in Advanced Driver Assistance Technology and Autonomous DrivingTowards Autonomous DrivingNew Design Wins 2017Main FeaturesTier 1Main FeaturesTier 1AEB EUNCAP 2018,LDWValeoAEB,LDWHirainAEB,ACC,LKAZF-TRWAEB,LDW,ACCHirainAEB,VOACC,LKAZF-TRWAEB,LDW,ACCMandoAEB EUNCAP 2020,Traffic Jam Assist
2、,Road ProfileAptivAEB,ACC,LKAHirainAEB,ACC,LKA,FreeSpceZF-TRWLDW,FCWHirainAEB,LKAAptivAEB,ACC,LKA,TJAHirainAEB,ACC,HLB,FreeSpaceMandoAEB,ACC,LKA,Lane ChangesDIASAEB,ACC,FreeSpace,Road EdgeKSSAEB,LDW,ACCHirainAEB,ACC,LKA,TSRNidecAEB,LKA,ACCHirainBase:L2/3 premium:L3/4NIOAEB,ACC,LKA,Lane ChangesKSSAEB
3、,VOACC,Glare Free HB,3D VD,REMZF-TRWAEB,LDW,ACCHirainAEB,pedal confusion,Enhanced LKAZF-TRWAEB,LDW,ACCHirainAEB EUNCAP 2020,Traffic Jam Assist,Road ProfileValeoFull EUNCAP2020 compliance,3D VD,FreeSpce,ObjectsValeoAEB,LDWAptivAEB,LKA,ACCZF-TRWAEB EUNCAP 2020&NHTSA,Road MagnaL3,surround,Road Profile,
4、REMAptivNearly 70 vehicle models,27 OEMs,30 design wins(in 2016 there were 12 design wins)30 Design Wins27 OEMs70 car models2017 ReviewOEM LaunchSpecial FeaturesTier 1GM CSAV2AEB(fusion),LKA,HLB,TJA,Super CruiseZF-TRWAudiAEB,LKA,HLB,RoadProfile,zFAS A8AptivFordAEB(fusion),LKA,HLB,TJAAptivHKMCAEB(fus
5、ion),LKA,HLBMandoPSA wave 2AEB(vision only),VOACC,LKA,RoadProfileZF-TRWNissanPropilot(vision only)launch in the USZF-TRWProgram Launches 2017Installed by 2017 year endGM Super CruiseAudi zFASNissan ProPilotEyeQ Shipped2017 Review24M EyeQs shipped to date2017:9M EyeQs02,2504,5006,7509,00011,250201420
6、1520162017Thousands2.7M4.4M6M8.7M2018 Program Launches15 programs to be launched during 2018 14 OEMs (4 of which are Chinese)4 programs with EyeQ4(12 additional launches starting from 2019)2 programs with Trifocal camera configuration ALL programs have full-feature bundles(high-end)New features laun
7、ched in 2018:3DVD Traffic Lights Detection and Recognition Advanced Road features:Semantic Free Space,Holistic Path Prediction REMStrategyPhilosophy:a single effortLevel-4/5 AutomationL2,L2+,L3derivativesModel for Safety Guarantees Decouple Sensing from Planning mistakes that could lead to an accide
8、nt RSS-a formal model of the human judgement of common-sense of Planning Using RSS to provide safety guarantees Economical Scalability Automating HD-maps through a crowdsourcing approach Controlling the explosive computational demands of Driving Policy(Planning)Scalable,workload-diverse and low-powe
9、r SoC together with powerful ATOM cpuDRIVING POLICYAutonomous Driving:Three PillarsSENSINGEnvironmental modelREM(Roadbook)Road Experience ManagementLocalization at high accuracy(10 cm)(Planning)Negotiating in a multi agent game360 awarenessDrivable PathsSensing Detecting Road-users and spatially-com
10、pact objectsVehicles,Pedestrians,Cyclists,Traffic Signs,Traffic Lights(mature technology in series production as part of ADAS evolution)Parsing the RoadwayLane marks,road edges,path delimiters,drivable pathsDrivable path(s)by redundancy of(i)sensing,and(ii)HD-mapNecessary building block for automati
11、ng HD-map construction(REM)Parsing Roadway to sufficient details for L4 is an open problemConventional approaches avoid Road-parsing and use only HD-mapNotes:Sensing Path Delimiters in urban environments Sensing Path Delimiters in urban environments Sensing Road users and Path Delimiters Sensing Hol
12、istic Lane CenteringRoad Experience Management(REM)Visual LandmarksVisual Landmarkson mapSparse 3D Dense 1DCrowd sourcedStrategic Value of REMLeveraging ADAS Introduce REM software on EyeQ for front-facing cameras(leverage existing real-estate in the car)Bandwidth of data from car to cloud is very l
13、ow 10kb per kilometer of driving The process for creating and updating maps is automatic.Volume of ADAS-enabled vehicles enable very low“time to reflect reality”everywhere,rather than merely in“geo-fenced”neighborhoods.REM introduces highly scalable“live”HD-map at low-costLeveraging Crowd-sourcingAu
14、tomationDensity of data sourcesBuilding Blocks of REMHarvestingcollections of roadway data(lanes,etc.)and landmarks to create RSD at 10kb/kmfusing all the RSDs in the cloud into a RoadBookusing RB and realtime detection of landmarks localize the host car in the RB at an accuracy sufficient for Polic
15、y and vehicle controlAggregationLocalizationREMRB data projected onto image space.Road edge,lane marks,lane center,landmarks(in Yellow).RB data projected onto Google Earth.REM 2017 Achievements Preparing harvesting for 2018 production programs(BMW,Nissan,VW)Preparing RB covering all Japan highways i
16、n cooperation with Zenrin and Nissan Cooperation with NavInfo and SAIC for bringing REM to ChinaDeals ongoing with OEMs for Harvesting 2019 and beyondDeals ongoing with OEMs for RB usage for L2+(new ADAS category)Aftermarket“Mobileye 8 Connect”REM supported and deals for 2018 deploymentMapping neigh
17、borhoods across the globe for supporting internal L4 development as a turn-key solutionREMMapping of Japan highways-with Zenrin/Mapbox/Nissan for L3 launch in 2019REMSensing alone(righthand image)cannot robustly detect the drivable path to enable safe hands-free control.The Roadbook data can bridge
18、the gap as localization is based on a high degree of redundancy of landmarks and is therefore robust.L2+front-facing camera+Roadbook A leap in ADAS L2 features(LKA/ACC).REMSensing alone(righthand image)cannot robustly detect the drivable path to enable safe hands-free control.The Roadbook data can b
19、ridge the gap as localization is based on a high degree of redundancy of landmarks and is therefore robust.L2+front-facing camera+Roadbook A leap in ADAS L2 features(LKA/ACC).Mobileye 8 Connect REM in the AftermarketVolume of device shipment to fleets 10%of OEM business(and growing),but each vehicle
20、 drives x10 the mileage of passenger carsA big data potential for REM Mobileye 5,6,7 EyeQ2 Shield+Mobileye 8 Connect EyeQ4+ModemLaunch Q1/2018Mobileye 8 Connect REM in the Aftermarketwhere pedestrians are more vulnerable to accidents and together with REM can provide data about infrastructure(lanes,
21、traffic signs)for decision makers Hotspots highlight areasMobileye 8 Connect REM in the AftermarketMobileye 8 Connect REM in the AftermarketMobileye 8 Connect REM in the AftermarketMobileye 8 Connect REM in the AftermarketDuring 2018Deals that have been signed with REM deploymentPartnerCity#of Vehic
22、lesGoal TimingKoMoD Research Project(Germany Ministry of Transport)Dusseldorf,Germany750Prepare the city for smarter&safer driving Q1Gett London 500Map City of London Q2Buggy New York 2,000Map City of New York Q3Road Safety Authority DGT(Directorate-General of Traffic)Spain Undisclosed,Spain 5,000Ma
23、ke Spain Autonomous ReadyQ2/3Guard InsuranceAcross the US 1,000-2,000 tow trucksQ3How to Guarantee Safety?AV Safety:the“Elephant in the Room”How to Guarantee Safety?Absolute Safety is impossible typical highway situationBeforeAfterHow to Guarantee Safety?“Self-driving cars should be statistically be
24、tter than a human driver”Problems:Not transparent:What will happen when a self-driving car will be involved in an accident?Will society be satisfied with the statistical argument?Infeasible:Theorem:to make sure that the probability of an accident per hour is at most p,one must drive more than 1/p ho
25、urs after every update of the software.can be an acceptable target(x1000 than human)RSS PrinciplesRSS Principles Self-driving cars should never causeaccidents Self-driving cars should properly respond to mistakes of other driversGoal:Self-driving cars should never be responsible for accidents,meanin
26、g:A mathematical,interpretable,model,formalizing the“common sense or“human judgement of“who is responsible for an accidentRSS is:Requirements for a formal responsibility modelSoundness:When the model says that the self-driving car is not responsible for an accident,it should clearly match“common sen
27、se“of human judgementUsefulness:it is possible to efficiently create a driving policy that guarantees to never cause accidents,while still maintaining normal flow of traffic.Covering all multi-agent scenarios in the NHTSA Pre-Crash Scenario Typology For Crash Avoidance Research AV Safety What is a s
28、afe distance in all situations What is a Dangerous Situation What is the Proper Response to it?Mathematically formalizing“human judgment”&common sense by defining:DefineGuarantee/Prove ExecuteOur approach:ResponsibilitySensitive Safety(RSS)AV Safety AV will never initiate a Dangerous Situation AV wi
29、ll always follow the Proper Response Namely-AV will never cause an accidentThis is how we can guarantee that:Creating a scalable and easy to validate platform Separating planning from sensing 109 105 Calculating one step ahead without“Butterfly Effect”DefineGuarantee/Prove Execute What is a safe dis
30、tance in all situations What is a Dangerous Situation What is the Proper Response to it?Mathematically formalizing“human judgment”&common sense by defining:Our approach:ResponsibilitySensitive Safety(RSS)AV SafetyCreate a scalable and easy to validate platformDefineGuarantee/Prove Execute What is a
31、safe distance in all situations What is a Dangerous Situation What is the Proper Response to it?Mathematically formalising“human judgment”&common sense by defining:AV will never initiate a Dangerous Situation AV will always follow the Proper Response Namely-AV will never cause an accidentThis is how
32、 we can guarantee that:Planning errors guaranteed by RSS Sensing validation is data-driven No“Butterfly”effect Separate“comfort”planning driven by Reinforcement Learning from“safety”driven by RSS.Our approach:ResponsibilitySensitive Safety(RSS)RSS PrinciplesRule TwoUnless the front car performed a r
33、eckless cut-inRSS formalizes common sense rules of determining -Rule Three Right-of-way is given,not takenRule One A hit from behind is not the front cars faultRule Four Be careful of areas with limited visibility“who is responsible for an accidentRSS PrinciplesA method for verifying that the AV tra
34、nsitions only between Safe States(notion of“Proper Response”).Concept of“Safe State”:a state in which an AV cannot cause an accident of its blame regardless of what other agents do.Formalize the“common sense”of human judgement in negotiating traffic(safety,legal and culture).Set the rules of“Blame”i
35、n advance.StrategyPhilosophy:a single effortLevel-4/5 AutomationL2,L2+,L3derivativesModel for Safety Guarantees Decouple Sensing from Planning mistakes that could lead to an accident RSS-a formal model of the human judgement of common-sense of Planning Using RSS to provide safety guarantees Economic
36、al Scalability Automating HD-maps through a crowdsourcing approach Controlling the explosive computational demands of Driving Policy(Planning)Scalable,workload-diverse and low-power SoC together with powerful ATOM cpuEyeQx Family:Terra OPs/W6 x VMP+2 x PMA+2 x PMC+4 x CPU,28nm,series prod from 3/201
37、8 launches by 4 OEMs in 2018,12 OEMs in 2019 and onwardsEyeQ30.25 TOPs 3W4 x VMP+4 x CPU,a 40nm,series prod since 11/2014EyeQ4H2.5 TOPs 6WNvidia Parker:1.5TOPs/15WEyeQ5HNvidia Xavier:30TOPs/30W7nm,1st silicon 8/2018,series prod from 3/2020 design wins by 4 OEMs from 2020 and onwards.24 TOPs 10WEyeQ5
38、Introducing:EyeQ5 technologyLPDDR4 PHYLPDDR4 Crtl&schedCacheCoherentInterconnectComputer visionprocessorsDeep Learning AcceleratorMultithreaded AcceleratorCPUsPeripheralsISRAMBoot ROML2 cacheL2 cacheLPDDR4 4267CPU clusterEst 52,000 DMIPS,with multithreading and virtualisation supportMultithread Acce
39、leratorMore versatile than a GPU with higher efficiency and utilisation than any CPUComputer Vision ProcessorsMarket leading Computer Vision Accelerators for computationally scalable sensor processingDeep Learning AcceleratorEst 24 DL TOPS2 10WSensor InterfacesIntel Atom SoC(DNV,XEON)EyeQ5(Vision)Ey
40、eQ5(Fusion&Policy)Cameras/sensorsTo actuatorsSENSEPLANACTFail Operational ChannelHarnessingThe Power of IntelEyeQ5 open to 3rd parties as an“open compute”platform with SDK and LibrariesIntelFleet:100 vehicles for testing,data collection,validation and customer supportIntelIntelData Center:250Pb for
41、supporting Fleet,validation and customer supportIntelEyeQ6 w/AtomSolutionArchitectureRadar/LIDARPartnershipsL3 Production(series development)2019+L4 Production(strategic partnerships)2020+L4 Turnkey solution Audi,BMW,Fiat-Chrysler,Honda,NIO,Nissan,SAIC BMW,Fiat-Chrysler,SAIC,NIO 3 x OEMs ongoing sou
42、rcing decisions CSLP platform with Delphi(Aptiv)Intel/Mobileye internal fleet of 100 vehicles ramping up throughout 2018 Master Plan2018L4 partnerships Turnkey solution:perception,driving policy,safety,MDC prototype(2 x EQ5+Atom)-platform derived from the 100-car fleet.Perception turnkey(EQ5)whereas
43、 Fusion,Driving Policy on open-EQ5(software as joint collaboration or solely by partner OEM/Tier-1).Open-compute+libraries:open-EQ5,Denverton,Xeon,Altera.L2+programs:Front-Facing sensing+Roadbook(“ADAS 2.0”)RSS with industry and regulatory bodiesREM as a“data strategy”IntelIntelIntelTHANK YOUDrive Safe!