20171016-汇丰银行-全球-科技行业-自动驾驶：通向AI之路的起点-HSBC-Nomadic_Investor_update：Autonomous_vehiclesthe_beginning_of_the_AI_road.pdf

Disclosures&Disclaimer This report must be read with the disclosures and the analyst certifications in the Disclosure appendix,and with the Disclaimer,which forms part of it.Issuer of report:HSBC Bank plc View HSBC Global Research at:https:/ MiFID II ResearchIs your access agreed?CONTACT us today?Autonomous vehicles(AVs)are set to mark the beginning of the autonomous age?The transition to AVs could be faster than many expect;we explore the lessons learnt so far from the migration to electric?We look at the core AV ecosystem and infrastructure,and highlight companies exposed to this likely new megatrend The autonomous revolution We believe that today society is on the cusp of the autonomous age where machines operate physical world objects that previously required human brainpower and manual dexterity,all through the power of modern Artificial Intelligence(AI),all by themselves and without supervision.The first big visually noticeable application of this trend is likely to be AVs because of the volume of key industry stakeholders already involved,readying the technology for mass market commercialisation on a global scale(see Global autos Disruptive threats:Carmakers versus new entrants,19 September 2017).In this report we build on our previous thematic work(see Transport shock:autonomous today,virtual tomorrow,19 October 2016)by looking at the ingredients of autonomous transportation,drawing parallels and lessons from another disruptive force within transport,the evolution of electric vehicles(EVs).Adoption of AVs could be faster than expected Rapid modern day advances in AI technology suggest that this new method of transportation could be adopted faster than many expect.Unlike the adoption of EVs,where progress has been limited(up until very recently)by battery capacity and associated high costs,the adoption of AVs is unlikely to face the same restrictions.This is because the pace of advancement in AVs is governed by computer technology(given Moores Law)and is likely to be much more rapid.Where to look for AI exposure We believe that now is the time for investors to understand the wider implications from AV rollout and also the new ecosystem required to support its growth over the coming years.Our Asia technology team has published a companion note called Artificial Intelligence:Chips with everything,16 October 2017,in which they take a deep-dive into the hardware required for AI and hence AVs.We highlight four themes within this AV ecosystem:computer vision,AV brain,connectivity,and automobile hardware(page 2).16 October 2017 Davey Jose*Thematic Strategist HSBC Bank plc +44 20 7991 1489 Ashim Paun,CAIA Climate Change Strategist HSBC Bank plc +44 20 7992 3591*Employed by a non-US affiliate of HSBC Securities(USA)Inc,andis not registered/qualified pursuant to FINRA regulations Nomadic Investor update THEMATIC GLOBAL Autonomous vehicles:the beginning of the AI road?THEMATIC GLOBAL 16 October 2017 2 Autonomous transportation ecosystem and infrastructureSource:HSBC ResearchComputer vision:Cameras,radar and LiDAR(Light Detection andRanging)Autonomous vehicle(AV)brain:High performance computerprocessors andAI softwareConnectivity:Enabling real-time dataexchange within AV systemSensors convert real-worldvisual information into data AIcan understandFusion of hardware and softwarecreates AI brain to process andunderstand visual dataAVs are essentiallydatacentres on wheels andcommunicate between otherAVs and infrastructureAuto OEMs and suppliers canpartner with new AV techplayersSensor ecosystem:3D Lumentum,Himax,Infineon,LG InnotekMotion Samsung LSIRadar Infineon,NXP,STMicro,Tung ThihLiDAR Velodyne,Delphi,QuanergyImage Sony,Samsung systemLSI,Omnivision,OnSemiconductor,PanasonicProcessing ecosystem:Autonomous driving platformsolution nVidia(Drive PX2),Intel(GO).Qualcomm(Snapdragon S820Am)CPU(central processing unit)nVidia,Intel,QualcommNPU(neural processing unit)Huawei(NPU in Krin AP),Apple(A11),Google(TPU),Qualcomm(Zeroth),Nvidia(Drive PX2),IBM(TrueNorth),Intel(Mobileye),AMD(radeon MI/RXVega),Samsung,Nepes(NM500)Memory Samsung,SK Hynix,Micron,Western Digital,ToshibaGeneral and auto AIframework TensorFlow,Caffe2,Theano,Torch,MXNet,Deeplearning4j.Keras/CNTK,BigDL/Open,NEO withOpenPilot(Comma.ai)Connectivityecosystem:5G modern developmentQualcomm(Snapdragon X50),Intel(5G Goldbridge),Samsung(5G RFIC),MediatekServer/datacentre HP,Lenovo,Quanta,IBM,Inspur,Oracle,Sugon,ChinaCacheCloud computing:Public cloud AT&T,BT,A,Telefonica,Cleversafe,Swisscom,SKT,Olleh KT,Softbank,Rackspace,VerizonPrivate cloud Openstack,Citrix,VMware,Microsoft,Redhat,HPHyperscale cloud Amazon,Microsoft Azure,SoftlayerAutomotiveecosystem:OEM Hyundai,BMW,GM,Benz,Honda,Volvo,AudiAutomotivesemi/componentsContinental,Bosch,Delphi,Honeywell,Valeo,Hyundai,Samsung,LG Innotek,SEMCO,Largan,Chin-PoonIndustrial Renesas,TexasInstrumentICT giants Google/Waymo,Baidu,nVidia,DeNA,SBDrive,Intel,Naver LabsMobility-as-service Uber,Lyft,Zipcar,Otto,DidiChuxing,OlaNew vehicle Tesla,BYD,Ninebot,StealthAutomobilehardware:?3 THEMATIC GLOBAL 16 October 2017?Sell the house.Sell the car.Sell the kids.Find someone else.Forget it.Im never coming back.Forget it.Apocalypse Now(1979)Riffing off a line from the 1979 motion picture,Apocalypse Now,directed by Francis Ford Coppola,we dont recommend one necessarily to sell their kids or other worldly assets.However,in the near future,it might be possible to sell your human-driven automobile and purchase(or ride-hail)a new kind of vehicle,one that makes your life easier,cheaper and safer:an AI-infused autonomous personal transportation vehicle.The idea is that once one embarks on the road to autonomous,one might never want to go back.Even before then,the world of personal transportation is already going through substantial disruption.Today,vehicles are migrating from the 100 plus year-old technology of the internal combustion engine(ICE)and transitioning towards modern day electrification.After electric vehicles(EVs),the next step,as we outlined in a previous thematic report,Transport shock:autonomous today,virtual tomorrow(19 October 2016),is widely suggested to be autonomy,facilitated by artificial intelligence.In that note we looked at the implications of autonomous transportation for jobs,leisure,consumption,infrastructure and how the changing nature of young and old demographics could be supportive of this new autonomous vehicle(AV)landscape.More recently,Horst Schneider,HSBCs head of autos equity research,together with Henning Cosman,published a deep-dive report called Global autos Disruptive threats:Carmakers versus new entrants(19 September 2017).In that report,they outline that carmakers can cope with challenges from EVs but they face more significant disruption from self-driving cars.Moreover,recent media coverage has highlighted that this autonomous future might be closer than one thinks.It has been suggested that Googles spin-off Waymo may be readying its own fleet of fully autonomous self-driving cars for public consumption,on the streets of Phoenix(Arizona),as soon this autumn.1 If this materialises,it could fuel and accelerate the AV competition,marking a significant inflection point for the future of transportation.The primary difference between AVs and EVs is that EV is to do with energy storage(hence chemistry)and AVs build out from computer technologies derived from Moores Law(computational ability essentially doubles but halves in cost every 18-24 months and has been driving our microprocessor fuelled technological advances over the last 50 years).So even though AV and EVs are both transport technologies,they differ in their foundational technologies(energy and arguably the laws of thermodynamics vs computational processing).Nevertheless,even taking these differences into account,we believe the transition we have seen so far to EVs can provide some useful indicators and lessons on the likely road to fully autonomous vehicles(Table 1)._ 1 Fully driverless cars could be months away,Ars Technica,October 2017.Introduction Sell the car.Go autonomous.Transport shock is already here Autonomous today The difference between AVs and EVs?THEMATIC GLOBAL 16 October 2017 4 Table 1.Autonomous read-across and open questions(vs EV lessons learnt)Theme AV vs.EV 1 Pricing systems Are extra electronic component costs easier to justify by autonomous convenience gained?Component costs likely to increase from level 1 to 5 but commoditise over time.Pricing systems from outright ownership,ride-hailing to open-source AV(autonomous vehicle)platforms.Is larger battery(vs ICE,internal combustion engine)cost justified for consumer?Prices likely to decrease with time though.2 Convenience Productivity time gained from not driving.Also from potential fleet traffic efficiencies.Charging time is bottleneck.However,this duration could fall for a full charge with technological improvements.3 Regulation&safety Safety is crucial.Early AV accidents could increase regulations and stifle adoption.On the flip side,good AV could lead to regulation to encourage AV adoption.Might an increase in travelling safety due to AVs foster pro-AV regulation like AV only lanes,taxing human drivers or tax breaks?Governments globally support EV(electric vehicle)in relation to climate change and cleaner cities to reduce carbon emissions.Various EV tax credits available globally.4 Climate change Roll-out of ride-hailing AV has unknown climate change implications today.Some suggest fewer vehicles on road due to ride-hailing AVs,others suggest easier access to AVs imply more vehicle miles travelled.Datacentres powering AV ecosystem could be green.Emission regulations in favour of EV.5 Social inclusivity Likely to increase independency for the elderly and immobile and save the state and families both time and money related to in situ care provision.Today the waiting time involved with charging might be more tiring for the elderly than the faster ICE refuelling equivalent.In the future charging could be faster.6 EM-DM bridge AI AVs may be able to learn more on the sometimes chaotic streets of EMs than DMs.Some EMs may be able to pass AV-friendly regulations faster than DMs.State-sponsored AV data sharing in some EM nations,eg China,could aid domestic AI AV systems to leapfrog their DM AV counterparts.Rapidly developing EM nation cities have an incentive to shift to EV to make them more pleasant cities.7 Timing&transition Timing is tricky but many AV participants suggest early 2020s to all the way to 2035 for fully AV.Iterative improvements from ADAS to level 5 autonomy might impact second-hand AV-less ICE/EV vehicles.Repeated inflection point has not come to pass.Is it different this time?Might impact second-hand ICE vehicle prices and inventory.8 Technology&infrastructure ADAS to level 5 autonomous require improvements and price commoditisation in:(a)Computer vision:sensors and components like cameras,radar and LiDAR(light imagine,detection and ranging).(b)AV brain:combination of high compute processors(eg CPUs,GPUs,TPUs,NPUs central,graphical,tensor and neural processing units)and AI software.(c)Connectivity and networks to make AV ecosystem communicate in real-time and have vehicle-to-everything infrastructure(V2X).Battery storage and charging are improving.Lots of charging stations in city and rural areas are required to reduce range anxiety.Is there a difference between DM and EM(eg China)?9 Suppliers&value chain Large sunk costs-with unknown payoff until fully AV.However,lower levels of autonomy(0 to 3)may help ease income stream.Full value chain unknown as technology is still in development and testing phase.Lack of legacy maybe an advantage to new AV players.Large sunk costs for R&D and production.Pay-off suggested to be within reach,more than previous times.Value chain a known quantity.Key today are volume and pricing.10 Maintenance&ownership OTA(over-the-air)software(s/w)updates,eg iPhone/Android mobile phones.Likely to reduce vehicle ownership within an AV sharing economy(eg ride-hailing).AI AVs could use advanced self-diagnostics for maintenance,ultimately leading to robots fixing robots,taking human labour out of the loop in the day-to-day operation of AV fleets,in the consumer,business and industrial space.EV likely to be highly software oriented,so also over-the-air(OTA)updates.Reduce ownership and sales of ICE,once price falls.Generally EVs have fewer parts so easier to maintain.Source:HSBC estimates?5 THEMATIC GLOBAL 16 October 2017 Autonomous now but what can AVs learn from EVs?We attempt to make some read-across assumptions for AVs(versus EV lessons learnt),following the below sub-themes:?Pricing systems and convenience offered One of the crucial bottlenecks for EV adoption is pricing,which depends on battery cost.As battery technology improves and production volume/capacity increases,EVs naturally become affordable.See Asia EV and Battery:how China is helping to crack the cost conundrum,6 April 2016).We outline three pricing models that could play a part in AV adoption:outright ownership,ride-hailing/sharing AVs(RAVs)and open-source(including crowd-sourced)platforms.Another bottleneck for EVs is the question of convenience,for example today it generally takes longer to charge an EV than re-fuel an ICE vehicle.We suggest the time saved when in an AV could offer increased productivity within society.?ESG and regulations Regulations have been supportive of the development of EVs,in relation to the climate change issues raised by ICE vehicles.We believe regulations could also be supportive of AV development and adoption,through its increased safety offered on roads.At the moment there are several moving parts to determine whether AVs could be a positive or a negative for climate change,so this remains an open question.Social inclusivity is another angle we look into.We suggest that AVs are likely to benefit the elderly and the demographic shifts globally(see An age-old question for a detailed demographic study,30 November 2015).This is in contrast to EVs,where the driver has to spend more time waiting for charging.This waiting could be tiring for the elderly versus quicker ICE re-fuelling today.?DM vs.EM Rapidly developing EM nations generally have an incentive to be supportive of EVs to combat pollution.We suggest that counter-intuitively,the less developed infrastructure of EM cities might produce better AI AVs than the DMs and that some EM nations might be able to pass AV-friendly regulations quicker than DMs.AVs could be a leapfrog technology for EM,just as wireless internet is potentially for them.For example in Kenya,mobiles have been instrumental in the widespread adoption of mobile payments(see Unbundling the City:Beyond urbanisation,2 May 2017).?Timing and transitioning issues There have been repeated false dawns for EV take-up in the past2 but today a number of factors are highly supportive of electrification,including regulation and technology.The main ques