An "Autonomy day" for Tesla investors provided a chance for the company's top brass to talk about the latest developments in its self-driving software and hardware. More importantly, the occasion also put a microphone in front of Elon Musk, and we all know what happens when you do that.
Here's a collection of typically outrageous and colorful quotes from Captain Moonshot, who stands to go down as one of this era's most fascinating innovators.
Tesla's new FSD (Full Self Driving) chips
Tesla has ditched Nvidia chips as a crucial part of its autonomous driving system, replacing them with an in-house design that can accept live video streams from all the car's many cameras, and process all that information in real time, alongside data from other systems like the radar, GPS, mapping system, IMU, ultrasonic sensors and more.
The new chip performs about four and a half times as fast as the Nvidia Xavier chip it replaces, and there are two of them on the new FSD computer being installed in Tesla's cars, giving full redundant backup if a chip fails. And Tesla's decision to design its own chips (they're manufactured in Texas by Samsung) has resulted in around a 20 percent cost saving for each digital car brain it builds.
Musk: "At first it seems improbable. How could it be that Tesla, who has never designed a chip before, would design the best self-driving chip in the world. But that is objectively what has occurred. And not best by a small margin, best by a huge margin. It's in the cars being produced right now. We switched over from the Nvidia solution for S and X about a month ago, and we switched over for Model 3 about 10 days ago.
"All Tesla cars being produced right now have everything necessary for full self driving. All you need to do is improve the software.
"We finished this design about one and a half, two years ago, and began the next design. We're not talking about the next generation today but we're about half way through it. All the things that are obvious for a next-generation chip, we're doing ... it'll be at least three times better than the current system. It's about two years away. I think if somebody started today, and they were really good, they might have something like we've got today in about three years. But in two years, we'll have something three times better.
"Nvidia is a great company, but they have many customers, so as they apply their resources, they need to do a generalized solution. We care about one thing: self driving. It was designed to do that incredibly well, and the software is designed to run on that hardware incredibly well. And I think the combination of that software and that hardware in unbeatable."
Self-driving computers and energy efficiency
Tesla has designed the FSD chip to be highly energy-efficient in its operations, targeting a power consumption no higher than 250 watts to make sure it doesn't impact range.
Musk: "It depends upon the nature of the driving as to how many miles [power consumption of the self driving computer] that effect is. It'll have a much bigger effect in the city than on the highway. If you're driving in the city and you had a solution that hypothetically uses a kilowatt, then you'd lose four miles on a Model 3. If you're only going 12 miles an hour, that'd be a 25 percent impact on your range in the city. Basically, the power of the system has a massive impact on city range, which is where we think most of the robotaxi market will be. So power is extremely important."
Tesla doesn't believe in LiDAR
The kinds of LiDAR sensors that power many of today's adaptive cruise control systems don't impress Elon Musk, who believes LiDAR adds a redundant set of data to a self driving system at a high cost that'll badly impact the bottom line of autonomous car manufacturers and customers alike.
Musk: "LiDAR is a fool's errand, and anyone relying on LiDAR is doomed. Doomed! Expensive sensors that are unnecessary. It's like having a whole bunch of expensive appendixes. One appendix is bad, now you have a whole bunch of them? You'll see."
Instead, the cars will use a number of different cues to determine the locations of objects in 3D space. These include inferring depth from multiple camera views – like the human visual system offers, as well as depth from motion – effectively comparing images from a single camera and taking its motion into account to see which items move by how much. The system also takes things like distance convergence of parallel lines into account. Effectively, the car is constantly building point-cloud 3D models of the world around it.
Tesla believes it's years ahead of the rest of the field
Musk: "The thing that's a powerful, sustainable advantage for us is the fleet. Nobody has the fleet. Those weights are constantly being updated and improved based on billions of miles driven. Tesla has 100 times more cars with the full self-driving hardware than everyone else combined. At the end of this quarter we'll have 500,000 cars with the full 8-camera setup, the full ultrasonics. Some of them will still be on hardware v2, but we'll still have the data gathering capability. A year from now, we'll have over a million cars with the full self-driving computer, hardware, everything.
"It's just a massive data advantage. It's like Google has a massive advantage in search, because people use it. People effectively program Google with their queries and their results."
Musk also confirmed that everything human drivers do in their Teslas is being used to educate Tesla's self driving network: "Essentially, everyone's training the network, all the time, that's what it amounts to. Whether Autopilot is on or off, it's being trained. Every mile it's driven, at hardware v2 or above, it's being trained."
On real-world vs. simulated miles
Other companies bashing away at the problem of autonomy believe they can reduce or negate Tesla's fleet advantage by running their hardware and software through huge numbers of simulated miles, but Musk doesn't think so: "We have quite a good simulation too. But it does not capture the long tail of weird things that happen in the real world. If the simulation fully captured the real world, well, that'd be proof that we're living in a simulation I think. It doesn't. I wish! The real world's really weird and messy, you need the cars on the road.
"In simulation, you're fundamentally grading your own homework. If you know what you're simulating, sure, you can solve for it. But the real world is very weird and has millions of corner cases. If someone can produce a self-driving simulation that accurately simulates reality, that in itself would be a monumental achievement of human capability. They can't. There's no way."
On the current performance of Tesla's Autopilot system
With the new hardware installed, Tesla's Autopilot is becoming increasingly capable at reading roads as well as traffic. Part of Tesla's tech demonstration involved showing how the cars can read cues from the vision that's coming in, and produce a 3-dimensional map of the road, including bumps, dips, slopes and whatnot.
It also keeps itself busy trying to work out what's around blind corners, building theoretical projections of what's beyond the line of sight with what Musk describes as astonishing prescience: "The network is predicting paths it can't even see with incredibly high accuracy. It can't see around the corner, but it's saying the probability of that curve is extremely high, so that's the path. And it nails it. You'll see that in the cars today, we're going to turn on augmented vision so you can see the lane lines and the predictions of the car in the video. It's kinda scary to be honest!"
On aggressive lane changes
One potential problem as self-driving cars begin to interact with human drivers is that they might be so conservative in their approach to lane changes that other drivers might refuse to let them in, a problem that'll be familiar to most any learner driver. Merging in heavy traffic can be quite a dance, with eye contact, body language and all sorts of visual pleases and thank-yous that Tesla recognizes its cars won't be able to mimic.
Musk has a potential solution for folk that refuse to back down in the game of freeway lane change chicken: "We'll probably have a mode that goes beyond Mad Max mode to LA Traffic mode. Mad Max would have a hard time in LA traffic, I think. We're being more conservative right now, but as we gain confidence we'll allow users to select a more aggressive mode. That'll be up to the user. There is a slight chance of a fender bender. Not a serious accident, but a fender bender. So it'll be a question of if you're OK with having a slight chance of a fender bender in freeway traffic. Which is unfortunately the only way to navigate LA traffic."
Tesla expects to have full autonomy and a RoboTaxi service online in late 2020
Musk: "There are three steps to self driving. There's being feature complete, then there's being feature complete to the level where we think the person in the car doesn't need to pay attention, and then there's a reliability level where we can also convince regulators that this is true. We expect to be feature complete in self driving this year. We expect to be confident enough from our standpoint to say that we think people don't need to touch the wheel or look out the window somewhere probably around second quarter of next year. And we expect to get regulatory approval, at least in some jurisdictions, toward the end of next year.
"That's roughly the timeline I expect things to go on. Probably for trucks, platooning will be approved by regulators before anything else. If you're driving a long-haul you could have one driver up front, and have four semis behind in a platooning manner, and I think the regulators will probably be quicker to approve that than other things."
Looking at a history of Tesla's achievements to date, he added: "All these things, we said we'd do them, and we did it. We're gonna do the RoboTaxi thing too. The only criticism, and I think it's a fair one, is that I'm not always on time. But I get it done, and the Tesla team gets it done. We expect to have the first RoboTaxis operating next year, with no-one in them.
"When things are on an exponential curve of improvement, it's very difficult to wrap our minds around it because we're used to working on a linear base. But when you've got massive amounts of hardware on the road, the cumulative data is increasing exponentially. The software is getting better at an exponential rate. I feel very confident in predicting autonomous RoboTaxis from Tesla next year. Not in every jurisdiction, because we won't have regulator approval everywhere. But I'm very confident that we'll have regulatory approval at least somewhere, by next year."
On hiring your car out through the Tesla Network
As well as moving to disrupt the ride share and taxi industries with its own JohnnyCab service, Tesla is also planning to let its customers put their cars to work when they're not using them. Any car with the right level of specification will be able to join the Tesla Network, which will let the cars roam around taking taxi jobs and offsetting their own cost with an income of their own. Users can choose who, where and when they're happy to have the car on call for others.
Musk: "If you own the car, you can add it to the Tesla Network, and Tesla will take 25-30 percent of the revenue. It's really simple. We'll update the Tesla app on your phone so that you can summon the car from wherever it's parked, or commit the car to the fleet.
"Typically a car will have around 10 to 12 hours of usage per week. But if you have a car that can operate autonomously, maybe it can operate for a third of the week or longer. Maybe 55-60 hours a week, maybe a bit longer. The fundamental utility of a vehicle increases by a factor of five. That's a massive increase in the economic efficiency of your transport solution.
"The fundamental message that consumers should be taking today is that it's financially insane to buy anything other than a Tesla. It will be like owning a horse in 3 years. I mean, fine if you want to own a horse, but you should go into it with that expectation. If you buy a car that does not have the hardware necessary for full self driving, it is like buying a horse. And the only car that has the hardware necessary for full self driving is a Tesla.
"People should really think about that as they approach their car buying. It's basically crazy to buy any other car than a Tesla. We need to communicate that better, and we will after today.
"The probability of the steering wheel being taken away in the future is 100 percent. Consumers will demand it. I want to be clear. This is not me prescribing a point of view about the world, this is me predicting what consumers will demand. Consumers will demand at some stage that people aren't allowed to drive these two-ton death machines."
Source: Tesla
