Topic: Autonomous driving
Reuters reports that Waymo had sought a billion dollars and a public apology from Uber in settlement talks over the lawsuit the companies are embroiled in. Uber apparently quickly dismissed those requests as unreasonable, which isn’t all that surprising, given that it’s still far from clear that Waymo has the evidence it needs against Uber as a company rather than merely against Anthony Levandowski around the stealing of LIDAR technology. It’s also a sign that Waymo is perfectly happy for the court case to continue and for it to continue to distract Uber at a time when the latter already has a lot on its plate including several legal actions and more.
The California DMV has approved rules to go into effect next year which will allow companies testing autonomous vehicles on its roads to do so without human drivers or traditional human driving hardware in cars. That’s a change to the existing rules, which had explicitly prohibited such testing and required human drivers. This appears to be at least in part a response to the increasing popularity of Arizona as a testing location for driverless cars due to its looser regulations, something which has presumably irked companies based in California which would rather test their technology closer to home. All of this could eventually be superseded by federal regulation being contemplated, which would override all state-level rules on this topic, but that’s still some way off and it makes sense for states with an existing commitment to allowing such testing to move forward in the meantime. As before, companies will have to register and report details of both their testing and any disengagements (human interventions) and accidents involving their cars.
via The Verge
Nvidia has announced Pegasus, a mini-computer which it claims is powerful enough to operate all the functions of a self-driving car while having roughly the footprint of a license plate, and which will become available in the second half of 2018. One of the key challenges with self-driving cars is that the computing power to run them is often so large and power-intensive that the unit often takes up much of the trunk of the car (see this Google search) and requires significant fuel over and above that required by the engine. Miniaturizing that processing power and making it more efficient is key to making autonomous driving a reality, albeit only one of several big challenges that must be overcome before that happens. Nvidia is arguably the current leader in providing the GPUs and related technology for these cars today, while others have taken the lead in sensors or connectivity relating to the cars, and this leadership has been a huge boon to Nvidia’s overall prospects and performance.
GM’s Cruise Automation unit has acquired Strobe, a startup which has been working on “chip-scale” LIDAR technology for use in self-driving cars. LIDAR is one of the big bottlenecks in autonomous tech development, both expensive and low-volume at present, with Velodyne currently the dominant supplier. As this Recode piece points out, GM is a bit more deeply invested in autonomous driving than most other legacy carmakers, having acquired Cruise itself as the “brain” of the system and also running various experiments of ride sharing and other services through Cruise and the GM Maven brand, and this acquisition extends that integration. My guess is that the technology was at a fairly early stage – the company seems to have just 11 employees – and it’s therefore unproven, though GM had an existing investment and may know something other potential acquirers didn’t, allowing it to swoop in at an opportune moment to take it off the market. Waymo and Uber, of course, are battling in court over the latter’s attempts to make its own in the image (or otherwise) of Waymo’s.
Alphabet’s Waymo subsidiary and chipmaker Intel have launched separate campaigns to promote autonomous driving technology. While Intel seems to be going it alone and focusing on TV ads with celebrities like LeBron James, Waymo has partnered with several safety and advocacy groups for its campaign, which seems more aimed at starting a conversation using the hashtag #letstalkselfdriving than pushing out its message via ads, at least for now. Waymo is an obvious company to be pushing the technology given that autonomy is its raison d’être and it has its own cars on the street in various markets, while Intel is clearly aiming for the same kind of indirect approach it took to its famous “Intel Inside” campaigns back in the day. These are, after all, mostly awareness campaigns at this point – there’s nothing any consumer could buy after seeing the efforts from either campaign, and most consumers aren’t even aware of regulatory efforts in this area yet either. But both campaigns are clearly aware of broad skepticism shown in recent surveys about autonomous driving and want to start the education process early. Waymo’s campaign is particularly focused on the accessibility and safety benefits and its partners – which include an organization serving the blind and another serving seniors. That gels well with the NHTSA stats I shared earlier today, which demonstrated again the potential safety benefits of a computer not prone to alcohol use, speeding, or distraction driving a car.
The US Department of Transportation and National Highway Traffic Safety Administration have released data on fatal motor vehicle crashes during 2016 (a fuller report is available here while the link below is to a summary press release). The total number of fatalities (which includes drivers and passengers in vehicles as well as pedestrians, motorcyclists, and cyclists) was 37,461, up 5.6% after a larger rise in 2015, but following a long decline in overall fatality rates from the 1960s onwards. As in prior years, what the NHTSA describes as “human choices” such as not using seatbelts, driving while drunk, sleepy, or distracted, or speeding, continued to be a major cause. Remarkably, nearly half of in-vehicle fatalities were among people not wearing seatbelts, and nearly a third of fatalities occurred where the driver was under the influence of alcohol.
One of the rallying cries of the autonomous driving movement is always that it should dramatically reduce these fatalities, which are arguably already very low at just over one fatality per 100 million miles. Given the contribution of human choices like alcohol use, speeding, and distraction to the totals, that seems likely to be true if autonomous technology can at least match the performance of human drivers on the fundamentals of driving. On the other hand, given that the vast majority of cars on the road will still be human-driven even once autonomous cars start arriving, things like increased seatbelt use (currently at around 90% of vehicle occupants) would make a much bigger difference in the near term.
The Information reports that Waymo is gearing up to offer the autonomous ride sharing service it previously announced sometime this fall (i.e. in the next month or two) but that its technology still has real problems dealing with some basic situations like left turns without human assistance. That’s a pretty fundamental problem and indicative of the state of autonomous driving even at companies as far along as Waymo is (generally further than others), and even in locales where it’s been testing for quite some time and therefore should have really good data. It’s not clear quite how Waymo is going to resolve that issue (neither making three right turns nor remote human control seem like workable long-term solutions). But bear this in mind next time you hear a car or tech company talk about imminent autonomous driving.
via The Information
Waymo has finally succeeded in getting the due diligence report on Otto which Uber commissioned as it planned to buy the company unsealed, allowing many details about what Uber knew and when to emerge. However, like earlier disclosures in the lawsuit, it mostly confirms that Anthony Levandowski, the executive at the center of the case who is nonetheless not named in the suit took documents and other information with him from Waymo, while not providing evidence that Uber benefited from that. Uber, meanwhile, continues to say that when it found out about the document haul Levandowski had, it ordered him to destroy it all and not bring it to Uber – proving that assertion false is Waymo’s biggest challenge. Of course, Levandowski as an individual might still have used that information in developing at least one version of LIDAR technology he worked on at Otto/Uber, but that would only cover some of the claims Waymo has made in the case.
Ford and Lyft have announced a partnership under which Ford cars will begin running as part of Lyft’s network, first with human drivers and eventually with autonomous technology doing at least some of the driving. This is just the latest in a series of deals Lyft has done around autonomous driving, with previous ones including Drive.ai, nuTonomy, and Waymo, while it also works on its own autonomous technology effort. Ford, meanwhile, has been very clear about the fact that it sees ride sharing as the initial application for its autonomous driving efforts, but of course doesn’t have a ride sharing service of its own to test it with – partnering with Lyft is one way to accelerate that effort while also learning things that could be applied to its own effort if it chooses to go that way. Ford is the second car manufacturer to partner with Lyft overall, with GM an investor and early partner, though that partnership has definitely seemed looser recently. This is the key thing with pretty much all Lyft’s partnerships: they could all turn into something really interesting, but none of them commits the companies to do anything specific over the long term, which leaves Lyft vulnerable to being left at the altar by its various partners if they decide to go in a different direction.
Baidu Announces v1.5 of its Autonomous Platform and $1.5bn Fund to Invest in Projects (Sep 21, 2017)
Baidu has announced version 1.5 of its Apollo autonomous driving platform with several new features and also announced a $1.5 billion (10 billion yuan) fund to invest in 100 autonomous driving “projects” over the next three years. All the detail in the press release is around the platform, the traction it’s gained, and the new features, which include obstacle perception, planning, cloud simulation, high-definition (HD) maps and
“end-to-end deep learning” capabilities. When the platform first launched, it sounded impressive on paper but in practical terms appeared to be rather piecemeal and unfinished, with many necessary components missing. The new features certainly fill some gaps but don’t supply all the missing pieces, and it’s likely that the platform still isn’t really ready for prime time deployment, especially outside of China where Baidu doesn’t have granular mapping data. The fund, meanwhile, is not detailed at all in Baidu’s release and it’s really not clear whether it will be a venture capital-style fund for investing in companies, or whether it will be more in the nature of grants supplied to companies using Baidu’s technology in some way. Either way, it’s a significant chunk of money in what’s already a very crowded and high-spending field.
CNBC reports that Tesla is using AMD “intellectual property” in its work on chips to power the autonomous driving systems in its cars. Though investors seem to have taken that as a sign that AMD is supplying Tesla with chips, the CNBC report doesn’t explicitly say that, but does quote the CEO of AMD foundry spinoff GlobalFoundries as saying it’s working directly with Tesla on chips, which may suggest AMD isn’t totally in the loop. The CNBC and other coverage has noted that former AMD chip engineers are now abundant at Tesla, though the company has used Nvidia rather than AMD chips in the past. It’s also interesting to see Tesla contemplating such an architectural shift when it’s claimed that the innards of cars it’s selling today based on its existing architecture are capable of running full autonomy in future. The idea of Tesla increasingly designing its own chips would certainly be in keeping with the work led by Autopilot head Jim Keller when he spearheaded the A-series chip initiative at Apple – companies truly serious about software need to design their own hardware right down to the chip layer, an idea reinforced by this week’s iPhone 8 chip performance benchmarks. But the news also makes clear how unsettled the chip vendor picture still is in the automotive space, with Intel clearly finally gaining some traction alongside others who have done better in the early running.
BlackBerry and Delphi today announced a partnership which will see the latter use the former’s QNX operating system as a secure foundation for its autonomous driving system. What’s not clear from either the press release the companies issued or the CNBC report linked below is what operating system Delphi’s platform has been built on until this point, because it’s not brand new and the company has been talking about releasing it to car manufacturers in 2019. At any rate, as far as I can tell QNX will join Intel and its Mobileye subsidiary as partners around the system, which focuses mostly on pulling in sensor data and making sense of it, rather than complete control of the car. QNX is already a widely used operating system within the car industry and BlackBerry has spent a lot of time hardening it and demonstrating its ultra-secure credentials since its acquisition several years ago, something that’s likely to become increasingly important as cars become more and more like connected computers. Investors clearly see the partnership as a boon for BlackBerry, whose shares rose quite a bit after hours today, but Delphi is only one of a number of manufacturers building similar systems for smaller car manufacturers, while larger automakers will likely mostly build their own. Further competition in this space will come from companies like Waymo, who will develop their own sensor and sensor fusion technology to go with their autonomous driving software and therefore offer something more like a complete package in time.
Waymo Uses Intel Chips for Autonomous Driving Technology (Sep 18, 2017)
With data centers a big exception, Intel has struggled to take a major share of most of the new chip technology markets that have emerged over the last twenty years, failing in mobile, tablets, wearables, and others. The automotive space has been another where it’s clearly been very serious – its Mobileye acquisition being the biggest sign of that seriousness – and yet has lost out to other big chip vendors including Qualcomm and Nvidia for some big contracts. In that context, I bet it’s been begging Google/Waymo for years to let it talk abbot the two companies’ partnership in powering autonomous driving technology, because it’s something of a coup. The two companies are now finally talking about that partnership in blog posts and coverage by TechCrunch linked below. Waymo has largely developed its own computing platform for self-driving cars internally but has apparently leaned on Intel chips almost from the beginning. There’s definitely some of the article here that feels overblown – talk of scale, for example, seems odd in the context of a fleet that currently numbers in the hundreds, while the idea that autonomy and self-driving “represents a significant portion of the chipmaker’s business” also feels off even with the inclusion of Mobileye. The words “car” and “autonomous driving” barely appear in Intel’s latest 10-Q, for example, and mostly in the context of that acquisition. But this is a big win for Intel, and one that’s remained quiet for a remarkably long time. It won’t by itself dramatically change Intel’s fortunes in this space, but it’s great validation that Intel is a worthy player given that Waymo is considered one of the leading companies in autonomous driving.
This isn’t huge news, and I think people who follow the transportation industry and autonomous driving technology closely would probably know this already, but it’s worth noting these comments from Alphabet subsidiary Waymo’s CEO on the timing of various applications of self-driving technology. He said at an event today that he sees trucking and ride sharing being the first applications for autonomy, and that either one might be the first to be commercialized at this point. That’s very much in keeping with the conclusions I’ve reached and what I’ve heard from various other industry players – the fact that trucking largely involves long distances and highways dramatically simplifies the driving task there and enables platooning of vehicles, all of which means it has a much clearer near-term return on the investment in autonomous technology than most other applications. Ride sharing, meanwhile, typically involves cars which have very high utilization rates versus private vehicles, and is often limited to a well defined geographic area, making the training and gathering of mapping data a more manageable task too. Of course, we still don’t know quite what the business model for either of these applications will be – whether a licensing of the technology, a direct participation or revenue sharing agreement for the ride sharing market, or something else.