Self-Driving Cars: Waymo and GM are FAR ahead of Uber…And Everyone Else
SEE LAST PAGE OF THIS REPORT Paul Sagawa / Tejas Raut Dessai
FOR IMPORTANT DISCLOSURES 203.901.1633/.901.1634
psagawa@ / email@example.com
March 29, 2018
Self-Driving Cars: Waymo and GM are FAR ahead of Uber…And Everyone Else
Robocab fleets are the endgame for autonomous cars. GOOGL, which will offer fully autonomous commercial service in metro Phoenix by year end, is well ahead, but the 2-3-year lead will not shut off opportunity for competitors. We expect the market to develop city-by-city, with municipalities promoting broad participation. Winning solutions will have strength in 3 major areas: self-driving technology (mapping and control systems), consumer reach (customer acquisition, service management), and fleet logistics (vehicle maintenance and repair, financing). GOOGL (self-driving value $75-100B) has begun to build its ecosystem with CARS and AN, but others must soon find their own partners. We see GM ($5-10B) as #2 with its self-driving tech and fleet management assets, but also a need for a consumer platform partner (e.g. AMZN, AAPL). BIDU ($3-5B) is offering its self-driving software to partners as an open source license – a strategy that could allow it to narrow the gap to the leaders. It also benefits from a broader, international testing regimen. Uber ($?) is in crisis. Its core business is existentially threatened by autonomous TaaS, but reports suggest MAJOR struggles. It may have to exit. Other possible TaaS tech platforms (Ford/ArgoAI, Zoox, Nutonomy, Aurora, etc.) are too nascent to handicap and likely years from having commercially viable solutions. We do not see ADAS systems for private cars as likely to deliver full autonomy for years.
- Robocab service starting now – years before general autonomy. Fleet-based autonomous TaaS will be a $100B+ global market within 10 years. GOOGL’s Waymo will be the dominant player, but other companies – building ecosystems to provide the technology platform, the consumer platform, and the logistics platform – will emerge as significant entrants. We have written of this extensively (http://www.ssrllc.com/publication/self-driving-cars-building-a-team-to-bring-taas-to-market/, http://www.ssrllc.com/publication/autonomous-cars-self-driving-ambition/). While Uber’s failures will bring scrutiny, Waymo is ready to launch safe, reliable and commercially viable robotaxi service in Phoenix. Key to this are extremely detailed, reliably comprehensive and scrupulously up-to-date 3D digital maps of the geofenced service area, which off-load much processing from the in-vehicle systems.
- GOOGL is well ahead. Waymo has most of the TaaS puzzle put together. The tech platform has been honed through a decade and more than 5 million miles of road testing, exposing the system to challenging circumstances on a private road course and via billions of miles of simulation. CA testing results show more than 5,600 miles between human interventions – almost all for situations that would not have resulted in an accident. It has the best mapping capabilities in the industry – by far. It has reach to consumers via its ubiquitous apps and its popular operating platforms, and the routing and payments capabilities to manage the service backend. It has signed capable partners (CAR, AN) for fleet logistics, and has orders with Chrysler and Jaguar for up to 40K vehicles.
- GM and AMZN would be a perfect partnership. GM is the clear #2 in self-driving tech. In CA testing, it is reporting 1,300 miles between interventions – 2 years behind Waymo, but FAR ahead of everyone else. It is primarily testing in urban San Francisco, exposing its system to challenging conditions that its lesser rivals (everyone but Waymo) have largely avoided. We believe plans to launch commercial service in 2019 are ambitious, but that the slow pace of Waymo’s city-by-city rollouts and municipal support for a competitive market will leave substantial opportunity for GM, despite GOOGL’s head start. Still, GM will need to build out an ecosystem, with customer reach and mapping, its primary needs. We believe AMZN, with its strong consumer platform, Alexa assistant, and interest in autonomous delivery, would be an ideal partner.
- BIDU’s open source strategy is intriguing. BIDU has made its Apollo self-driving software available as a free open source license, drawing dozens of partners across 3 continents, intending to monetize its investment by selling access to detailed 3D digital maps as a service for its licensees. With this strategy, BIDU can collect a greater quantity of more diverse testing data to improve its system, while hoping to establish Apollo as a standard amongst OEMs – many of whom are likely to fail with their own home-grown solutions. BIDU also has a substantial advantage in China, where foreign companies have been excluded from testing. We believe that leadership in localized, detailed 3D digital maps will be a very valuable asset with substantial competitive barriers.
- Uber in crisis. Uber’s self-driving car SHOULD have been able to avoid the recent fatal pedestrian accident in Phoenix. This is the culmination of a long series of legal problems, organizational failures and poor performance that could soon end a development initiative that former CEO Travis Kalanick once cited as “basically existential” for the company. With reports that Uber vehicles travel just 13 miles between disengagements and credible stories of cars running red lights, freezing with indecision for common circumstances, and delivering uncomfortable rides (e.g. hard breaking, jerky steering, etc.), the company is years behind its rivals and its development team is dispirited. Ending the 4-year effort and joining with a technical partner (e.g. GM? BIDU?, a startup?) may be the only real option.
- It’s getting late for everyone else. Uber’s struggles illustrate the difficulty in bringing self-driving to market. It started with a crack team of experienced CMU roboticists, enthusiastic support from management, and time to log 2 million test miles, yet it has failed. The claims of others looking to jump into the TaaS market (e.g. Ford/Argo.AI, Lyft/Nutonomy, Aurora, Lyft/Drive.AI, Zoox) are based on small samples of driving data, and while one or more may emerge as a real player, all of them trail GM’s testing trajectory by many months and Waymo’s by years. It is also possible that Uber’s tragic accident will lead to greater road testing restrictions for companies at earlier stages of development. We are also skeptical that the many companies developing automated driver assistance systems (ADAS), like Tesla’s much hyped highway Autopilot, will be able to offer fully autonomous operation for many years.
- GOOGL and GM valuations do not reflect the TaaS opportunity. We believe that Waymo would be valued at $75-100B as stand-alone company, based on the potential for $10B+ in annual revenues by 2025. GOOGL’s current market cap of $699B, 23.9 times forward earnings, does not appear to ascribe value to the venture (or to GOOGL’s other growth initiatives, such as its cloud platform). Similarly, GMs Cruise Automation, bought for $1B in 2016, is likely worth at least 5 times its purchase price after delivering such promising results. Private companies that are far behind GM, such as ArgoAI or Nutonomy, have been receiving $1B+ valuations in recent funding.
Wheat vs. Chaffe
On the night of March 18th, an autonomous test vehicle, operated by Uber and under the supervision of a safety driver, struck and killed 49-year-old Elaine Herzberg as she crossed a four-lane road at mid-block. While initial police reports suggested that the victim entered the poorly lit roadway too quickly for a human driver to react, an autonomous car equipped with both LiDAR and RADAR should have easily identified her with plenty of margin to safely avoid any contact. Rather than an unavoidable tragedy or an indictment of all autonomous testing, this is serious failure of Uber’s entire self-driving program, which has stumbled through a legal challenge from GOOGL, embarrassing news flow at many levels of the company, and the loss of key engineering leaders, into existential crisis. According to the NYT, Uber’s autonomous vehicles manage just 13 miles between necessary human interventions – by contrast, GOOGL’s Waymo averages more than 5,600 miles, and GM, more than 1,300. Anecdotal reports from riders suggest cars that are flummoxed by common driving situations, that occasionally run red lights and that offer a choppy, uncomfortable ride. The Governor of AZ has suspended Uber’s license to test in the state. It seems that Uber’s once highly touted initiative may be, to paraphrase Ralph Nader, “unsafe at any speed”.
Playing Gallant to Uber’s Goofus, Waymo’s leader John Krafcik asserts that his cars would have stopped, and we believe him. With 10 years and 5 million miles of testing on public roads, combined with extensive recreations of challenging scenarios on its private testing grounds and many billions of iterations in simulation, reports on Waymo’s safety are impeccable. While it reported that 5,600 miles per disengagement ratio to CA regulators, further comments suggest that most of the interventions were initiated by the engineering team to log unusual events and that the vehicles could have otherwise managed safely. In this context, Waymo’s plan to offer fully autonomous commercial robocab service in Phoenix is sober.
GM’s Cruise Automation self-driving unit leadership has not commented on Uber’s accident, but it seems clear that it too can be confident that its vehicles would have safely navigated a similar situation. Its strong performance statistics were logged with a heavy dose of urban driving in hilly San Francisco, with no reports of reckless behavior, like those that have occasionally dogged Uber. It hopes to offer its own commercial robocab service by the end of 2019, but we suspect that this timeframe may be slightly ambitious.
Others are further from readiness. Of these, the most intriguing is BIDU. It is offering its self-driving solution as a free open-source license, with the expectation that its licensees will contribute testing data back to make it better. With this, BIDU has gained partners on 3 continents, while the other serious rivals are focused on the US. BIDU plans to monetize its investment by offering its 3D digital maps as a subscription. While the most recent test data posted in CA is not yet competitive, this approach could make up ground and positions the company as an early leader in China and other international markets.
We believe that the global autonomous TaaS market will exceed $100B within a decade, with Waymo the clear leader and GM and BIDU the best challengers. We believe Waymo would be worth $75-100B today as an independent company, and that GM’s Cruise Automation subsidiary might fetch $5-10B. None of this value is captured in the market caps of the parents, making both GOOGL and GM intriguing plays on the future of self-driving. ADAS systems for private vehicles – under development by car makers (TSLA, GM, F, BMW, Daimler, Nissan, etc.) and by auto suppliers (Bosch, Delphi, Denso, INTC/Mobileye, etc.) – will have a growing market, but are many years from enabling full autonomy.
How’s it Gonna Be
With Waymo poised to launch the commercial fully autonomous Transportation-as-a-Service (TaaS) operation in Phoenix, ride hailing robocabs will be first real market for self-driving technology, beating autonomous private cars, self-driving trucks and robotic delivery vehicles to market, likely by several years. We have written of this extensively (http://www.ssrllc.com/publication/googl-waymo-is-worth-75b-right-now/, http://www.ssrllc.com/publication/autonomous-trucks-self-driving-convoys-are-years-away/, http://www.ssrllc.com/publication/self-driving-cars-building-a-team-to-bring-taas-to-market/, http://www.ssrllc.com/publication/autonomous-cars-self-driving-ambition/). Autonomous TaaS is possible today because operations can be limited to a specific electronically geofenced area, where extraordinarily detailed, reliably comprehensive and scrupulously updated 3D digital maps can augment the sensor suite on board the vehicle to provide the best possible picture of the cars environment from which to make driving decisions. This frees the car from reading signs, measuring curb heights, choosing appropriate drop off points, identifying construction zones, and other static features of the environment in real time.
Extensive on-site testing also allows the self-driving system to be well-adjusted to local driving habits. For example, Arizonans, unaccustomed to precipitation, drive much more cautiously in the rain than drivers from wetter climates. Similarly, drivers in the northeast have a deserved reputation for driving more aggressively than those in the Midwest. The contrasts will be starker across international borders. A robocab trained on American driving habits would be completely unprepared for the conditions many cities overseas. While a generalized driving intelligence equally adept on the streets of New York, London and Bangkok may be possible, it is not necessary for the spread of autonomous TaaS and almost certainly many, many years away (Exhibit 1).
Exh 1: Elements in Phased Rollout of Self-Driving Technology
This is the big reason that robocab TaaS will be the first application for full self-driving, and likely the only application for some time. Google can afford to map and test the Phoenix metro area confident that it can leverage that investment with a big fleet of robotaxis providing hundreds of thousands of rides a day within that geofenced territory. Add in the high cost and obtrusiveness (roof top turrets) of the LiDAR sensors needed to provide safe self-driving operations under less than ideal conditions, and the practicality of privately owned vehicles with fully autonomous capability is even more in question.
We expect the market for autonomous TaaS to be huge. The all-in cost of personal transportation is a $1.23T budget item for American households and a $5T market worldwide (Exhibit 2, 3). 63% of personal driving is to familiar destinations within 5 miles of home. The average US driver spends 25 hours per month behind the wheel. With driver compensation making up more than half of the cost of ride hailing, with the potential of MUCH higher asset utilization, and with the advantages of an electric fleet, we expect Waymo to be able to undercut Uber’s current service price by as much as half and perhaps more – robots don’t expect tips. We believe this can sharply accelerate the already strong growth in ride hailing as robocab service rolls to 5-10 new markets over the next 5 years. We believe autonomous TaaS can be a $100B+ market by 2030 in the US alone (Exhibit 4).
Exh 2: US Spending on Vehicular Ownership and Maintenance
Exh 3: US Vehicular Trips by Purpose Type
Exh 4: US Autonomous TaaS Market Forecast, 2018 – 2030
Building an Ecosystem
A successful TaaS ecosystem will need excellence in three different business areas (Exhibit 5). First, and most obviously, is Self-Driving Technology. The tech role comprises two different elements, mapping and control. Mapping begins with static digital maps – as we noted above, the more detailed, comprehensive and up-to-date the maps, the better. The second aspect of mapping takes place on board, where a powerful AI fuses the inputs from a variety of sensors – LiDAR, RADAR, cameras, ultrasonic (for proximity), microphones, etc. – with the static maps to create a 3D picture of the world around the car, overlaying predictions about what all the moving parts are likely to do next. Finally, the control system is an AI that takes that information and makes decisions on how best to proceed.
Exh 5: Key Capabilities of Transportation as a Service (“TaaS”) Providers
Second, a TaaS ecosystem requires Consumer Reach, a role that comprises several related responsibilities. Customer Acquisition is obvious, but difficult. We believe this task will be made more difficult as a new generation of comprehensive AI infused user interface platforms – incorporating but not limited to voice commands – that will begin to squeeze stand alone apps for engagement. A user may simply ask for “a ride” and depend on an AI assistant to work out the details. We expect the big consumer cloud franchises – Google, Apple, Amazon, and perhaps Facebook – will have substantial advantages over the transportation specific services like Uber or Lyft. Customer Engagement involves the nitty gritty of taking requests, delivering rides and assuring user satisfaction. This has been the biggest role of the ride hailing pioneers – Uber, Lyft, and others – but one that may be easier to manage once the drivers are removed from the equation. For robocabs, engagement will also involve in-vehicle communications – we expect that ride monitors to handle requests and project service safety will be a key part of the experience. Finally, the consumer role must also handle the back-office Service Management. Agreements with municipalities must be forged and followed. Payments (we think service will shift to subscriptions to avoid ride-by-ride rate arbitrage) will need to be processed. Insurance will be dealt with, and claims handled. The usual HR, finance, etc. functions will be handled.
Finally, Fleet Logistics must be managed. Vehicles must be designed and manufactured to spec. These cars must be maintained to consumer standards – cleaned, charged/fueled, inspected, etc. – on at least a daily basis. If there is an accident, the vehicles must be repaired. Last, but not least, as the asset value of the fleets expands into the billions of dollars, the cars will need to be financed and held on someone’s balance sheet.
As companies move to enter the emerging TaaS market, it will be important for all these roles to be filled competently. Only Waymo seems to have begun to build a real ecosystem.
The Latest on Waymo
We wrote recently on Google’s Waymo (http://www.ssrllc.com/publication/googl-waymo-is-worth-75b-right-now/), laying out its leadership in both mapping and control systems, and its potential as a consumer platform. It has already recruited Avis and AutoNation to fill important gaps in its ecosystem as it moves to offer commercial TaaS in Phoenix. A recently announced deal with Jaguar for 20,000 of its new iPACE plug-in electric luxury SUVs to be kitted out with its autonomous driving package will augment a prior order for a similar number of Chrysler Pacifica hybrid minivans. CEO John Krafcik put it, the fleet would be able to complete a million trips per day by 2020. This is enough for Phoenix and to launch a handful of new markets to come (Exhibit 6).
Exh 6: Waymo Progress Capture on Self Driving Vehicles, March 2018
We do not believe that the controversy created by Uber’s tragic accident will hinder Waymo’s progress. Having begun testing self-driving solutions six years before any of its rivals, Waymo has had the luxury of time and has proceeded cautiously. Krafcik has expressed strong confidence that his vehicles would not have come close to striking the pedestrian killed by Uber’s car, and has the safety data from 5 million miles of road testing to back him up. Waymo has a private facility where it has recreated road configurations that have challenged its cars and run live tests with professional drivers. It has run billions of miles in simulation running through those challenging situations with changing parameters. Waymo is ready now.
In addition to Phoenix, Waymo has been testing in Silicon Valley, Austin Texas, Kirkland Washington, suburban Detroit, and Atlanta. These are likely candidates for expansion once the initial launch proves successful. Still, we expect Waymo’s progress to be methodical – from the start, Google has stressed safety, to the point that the former head of the initiative, Chris Urmson, left two years ago in frustration at the deliberate pace toward commercialization. In the wake of the Uber accident, that caution is proving an asset.
GM on the Move
When GM bought Cruise Automation in March 2016 for a reported $1B, no one was quite sure what they had bought. Cruise was one of several self-driving startups with excellent technical pedigree but no real public evidence of progress. Two years later, the deal looks like a steal.
Exh 7: GM Cruise Self-Driving Performance Highlights, 2017
In 2017 California road testing, GM’s autonomous fleet logged 132K miles and 1,300 miles on average between human interventions (Exhibit 7). These stats were second to Waymo (which had posted similar results back in 2015 and has improved since then) but were far ahead of every other company. Number three was Nissan, which managed less than 300 miles between disengagements. GM’s accomplishment is even more impressive, given the company’s high degree of difficulty – much of its testing was done on the urban streets of hilly San Francisco, which requires more complex driving decisions than Phoenix. Despite that, reports on the ride quality and attention to safety of the autonomous Chevy Bolts have been excellent. A 12 MPH fender bender with a motorcycle in January of this year received press attention, but the GM system does not appear to be at fault – it had aborted a lane shift as another car braked unexpectedly and the motorcyclist had accelerated between lanes anticipating the shift.
GM has announced that it will produce a vehicle designed for full autonomy, without a steering wheel, brake pedal or other driver controls, by 2019, with the aim of offering TaaS robocab service in the same year (Exhibit 8). This may be ambitious, but even a modest delay from this schedule would place GM on track to compete effectively with Waymo as autonomous fleet services slowly penetrate municipal markets across the US. Given the state of rival systems and the potential that scrutiny of Uber’s recent accident could lead to limits on testing for companies further back, this could be a two-horse race for many years.
Exh 8: GM Cruise Self-Driving Initiative Timeline