Teslas will be fully autonomous next year. The word, from Elon Musk, came just two days before a Tesla earnings-and-sales report that is not going to be kind to the company. The Tesla CEO and three Musk-eteers took to the stage Monday to announce fabulous, industry-leading breakthroughs in full self-driving (FSD) that will put Tesla cars and Tesla robo-taxis on the road as early as next year, in at least one market. The plan is to use two Tesla-designed chips that will fit in a glovebox circuit board rather than take over the trunk. All without lidar.
Musk said Teslas will be "feature complete" for self-driving as of vehicles built this year; self-driving (Level 5, the real self-driving) could be a reality by Q2 of 2020. (Self-driving presidential campaign vehicles, anyone?) That puts Tesla well ahead of every other automaker if Tesla, in fact, delivers what it promised Monday.
Peter Bannon, VP of Tesla Autopilot engineering, and Elon Musk at Monday's Autonomy Day.
A Breakthrough, or More Silicon Valley Hot Air?
If we sound a bit cautious, we've been there before with Tesla. Other Tesla promises have come up short: start-of-production claims, production-quantity claims, technology. And yet, Tesla is by far the largest maker of EVs, this from a company that didn't exist 15 years ago. And the Tesla Model 3, even if it failed to meet Tesla's delivery and production claims, still was the best-selling luxury car in the US last year and outsold the next EV, the Nissan Leaf, by 8-1.
Monday's event ran almost two hours, plus a delay of about 50 minutes past the scheduled 11 am PDT start time Monday. The three speakers in addition to Musk made it something of an academic, technical proceeding. But then self-driving is also more complex than a Hallmark movie. The speakers:- Peter Bannon, a former engineer with Apple PA Semiconductor. He detailed Tesla’s custom-built self-driving computer and chip, which places two independent FSD systems on a single board. "It's a pretty small computer, fits in the glovebox, between the glovebox and the firewall and does not take up half the trunk," he said. For those who believe technology commerce observes political borders, the custom processors are being fabbed by Samsung (boo! foreigners) in Texas (yay! America, and a red state no less).
- Andrej Karpathy, senior director of artificial intelligence for Tesla. He spoke on how Tesla neural networks collect data and how Tesla uses the data to inform the decisions its neural network makes in deciding edges of the road, pedestrians, signs, and the back ends of moving boats on trailers that signify they're a moving vehicle, too. "There is no substitute for real data," Karpathy said, noting Tesla has an advantage from data collected from existing Autopilot Teslas.
- Stuart Bowers, Tesla’s VP of engineering, formerly with Snap, Facebook, and Microsoft. He spoke on Tesla's self-driving software and said new features are first released in "shadow mode," in controlled deployment, giving Tesla more feedback. For example, Bowers said, Teslas make 100,000 auto lane changes every day, so far with zero accidents, and that data further improves the feature.
In their entirety, the presentations were on-point, thoughtful, and thorough. But it's just not possible to determine if fully redundant FSD systems on a single board, a high-speed neural network, radars, eight cameras, and 12 ultrasonic sensors are going to make Tesla fully self-driving in 2020. If Tesla did what it says it's doing, this is a quantum leap over the competition.
According to Bannon, "The [two separate cores] boot up and run their own operating systems. The two machines exchange … independent versions of [driving] plans and make sure they’re the same, and assuming that they agree, they act and drive the car." Musk added, "Any part of this could fail, and the car will keep driving."
Every Tesla now comes with Autopilot standard. Tesla says the model you buy today will run the more autonomous software available circa 2020. But then, back in October 2016, Musk claimed, "All Tesla cars being produced now have full self-driving hardware." That was an overstatement; Tesla later said it would need new hardware, and Tesla is on the hook for those new drive computers.
Here's one feature you may like if you're in a Tesla, but less so if you're driving alongside one, and you'll probably freak if you work for an agency with "Transportation Safety" in its name: Tesla plans a Mad Max setting that will drive more aggressively on highways. One example Musk gave is more aggressive lane changes.
Although Tesla is gung-ho on self-driving, however, Musk has nothing new to say about his 2017 promise that Tesla would soon be making an autonomous, coast-to-coast road trip.
How aggressive is Tesla's timetable for Autopilot FSD (previously Autopilot 3.0)? Very. Not only is it next year, but it's also said to be "anywhere" technology, and done without lidar sensors that draw detailed surroundings-maps even in mixed weather. In comparison, the CEO of Ford, Jim Hackett, said the industry misunderstood the complexity of self-driving. Ford is still sticking with 2021 as the year it launches self-driving, but only in certain areas.
Tesla FSD (full self-driving computer) has two redundant processor cores, each with its own DRAM (let and right of the core), flash memory (below-left of core), and separate OS. At right on edge, 8 video camera inputs. It fits between the glovebox and firewall. (The color is highlighting from the Tesla presentation. Don't look for it on production boards, okay?)
Robo-Taxi Fleet Makes Your Tesla Worth $200,000
Monday, Musk paired his bold timeline for self-driving with a claim that the net present value of an individual's Tesla put into robo-taxi service would be around $200,000. "The fundamental message that consumers should be taking today is that it’s financially insane to buy anything other than a Tesla," Musk said. "It will be like owning a horse in three years—fine if you want to own a horse, but you should go into it with that expectation." A bold promise, and it may have some wondering if a Tesla worth $200,000 is what the Securities and Exchange Commission would call a forward-looking statement. Musk says there could be 1 million robo-taxis up and running in 2020. (New York City has the largest traditional taxicab operation, and that's just 13,000 cabs, although now outnumbered by Uber, Lyft, and others.)
Here's the idea: You buy a Tesla, with Autopilot, that's fully self-driving. You put it into a Tesla autonomous drive taxi fleet, using software developed by Tesla. You make a bunch of money off the taxi service, maybe $30,000 a year. Tesla gets a cut, say 25 to 30 percent. This raises the value of your car because it has a secondary use beyond driving you around. The taxi miles don't really affect your Tesla's longevity, Musk says, because it's good for 1 million miles of driving. In areas where there aren't enough Teslas to make the robo-service viable, Tesla would supplement the fleet with its own taxis. In other words, an area where there aren't many Tesla owners but there are people who'd pay to ride in a Tesla, autonomously, over Lyft or Uber.
Will it really happen? Musk said he's "very confident" that it will launch in some areas, but not all because of local or state regulations. Speaking more to Wall Street more than Tesla owners, Musk said deploying such a Tesla-owned fleet would have no material impact on the company's overhead, costs, or earnings. He may well be asked about that at Wednesday's earnings call, slated for after the market closes.
For those who believe cars are meant to be driven, and paranoid about human-driven cars being outlawed at some point, Musk gave them something to worry about: "I’m very convinced ... [that] in the future, people will want to outlaw people driving their own cars because they’ll be unsafe" relative to autonomous vehicles.
Musk also further annoyed Nvidia, maker of self-drive components for several automakers, with the statement that Tesla has "the world’s fastest computer" for autonomous driving, able to run 144 trillion operations per second (TOPS). Nvidia fired back quickly, said Tesla made an oranges-to-former-Apple-employees comparison, and the Nvidia self-drive processor runs 320 trillion operations per second. Almost enough to make Photoshop seem fast. Nvidia's statement:Tesla was inaccurate in comparing its Full Self Driving computer at 144 TOPS of processing with Nvidia Drive Xavier at 21 TOPS. The correct comparison would have been against Nvidia's full self-driving computer, Nvidia Drive AGX Pegasus, which delivers 320 TOPS for AI perception, localization and path planning.
Tesla's self-learning that is relayed back to Tesla and then broadcast to other cars includes identifying and locating debris on the road. (For those living outside sunny California, pothole recognition would be nice, too.)
Tesla Media Coverage Grows Skeptical
You'll likely read mixed reporting and analysis of Monday's event. Much of the media and analysts see Musk as part genius, part risk-taker, part PT Barnum. What he says today may not pan out tomorrow.
Thus their reports:Elon Musk, Tesla’s chief executive, has made plenty of bold predictions. They don’t always come true. On Monday, Mr. Musk said the company was on the cusp of making cars that could drive themselves safely on any road. He also promised that the company would begin operating a fleet of driverless “robo taxis” by the end of next year. ... Many auto executives and analysts think Mr. Musk is being wildly optimistic and say cars that can drive themselves at all times are at least several years away. - Neal E. Boudette, The New York Times Tesla once again makes the same Level 5 autonomy promise that it made in its October 19, 2016, press conference. Once again, CEO Elon Musk said that the current Tesla cars have the hardware necessary to operate at Level 5 autonomy, with just a software update. Well, since October 2016 nothing happened for Tesla beyond Level 2 Autonomy, and what they did do was not one but two hardware updates! Tesla now caveats its 2020 Level 5 service promise by saying it’s dependent on regulatory approval. However, Tesla fails to specify what approval is necessary. - Anton Wahlman, Seeking Alpha
Mr. Musk’s predictions come as Tesla prepares on Wednesday to reveal its first-quarter financial results, which are expected to show a loss on slumping sales, raising questions about the demand for the company’s Model 3 compact car. The quarter, which included a large debt payment, likely ate into the company’s limited cash balance. Analysts say Tesla needs to raise billions of dollars to fund the company’s growth plans, but Mr. Musk has spoken out against raising money for the past year. - Tim Higgins, The Wall Street Journal Karl Brauer, executive publisher of Kelly Blue Book, is skeptical of the aggressive timeline Musk and company presented, especially in light of regulatory and safety concerns about fully autonomous driving. “Clearly he and about 20 other entities are going to have (full) self-driving (vehicles) someday,” Brauer said. “If you’re a couple of days before an earnings report and you’re trying to make sure investors are engaged and excited about your company, this is a way to do that.” - Levi Sumagaysay, San Jose Mercury News
The upshot on Monday's Autonomy Day, as Tesla billed it: Tesla realizes it needs to do a better job reaching out to analysts and the media to explain the technology behind Tesla in general and Autopilot in particular. Tesla stock is down 15 percent this year and 32 percent from its September 2017 peak. Analysts are always worriers. At the same time, buyers need reassurance that they made a good choice. And most of all, Tesla needs to convince the 98 percent of the public that didn't buy EVs last year that there's a solid future for electrification. It's an easier case to make when gasoline costs $4 a gallon rather than the current $2.85.
Stay tuned for the results of Wednesday's analyst call. The news will not be good. US sales collapsed when Tesla's EV tax credit got halved Jan. 1. Meanwhile, keep thinking about whether Tesla has made a great leap forward in technology, or in marketing spin. Or some of each.
Now read:
