Video: "Tesla Stanley" explains (badly) why Tesla FSD rules self-driving

Topic: 

It's surprisingly common to see articles and videos from Tesla fans who believe that Tesla FSD is a contender -- even the leader -- in self-driving development. The reality is it's still not even in the race compared to other players. In this video, I pretend to be a super Tesla FSD fan and I give the reasons they commonly give for feeling that Tesla rules the roost, but in a way that makes the flaws in those arguments more apparent.

This is best watched as a video, but for those who want to read the script, here it is below:

This is Brad Templeton from Robocars.com, but today I'm going to take on a new personality. Today I'm "Tesla Stanley" and Stanley is here in my Tesla to tell you howTesla FSD is the leading self-driving project in the world. Forget Waymo, Cruise, MobilEye, Gatik, Zoox, Baidu, AutoX and all the others -- I'll tell you why the Tesla approach is going to win. I bought Tesla FSD for my car back in 2019 and have been using it since 2021, and I'm going to tell you the scoop.

The latest release of Tesla FSD has improved immensely since the early days, though of course it was incredible and great back then too. Each new release in the release notes tells us how they've done things like increased precision of detection of vulnerable road users by 44.9% as they did with 10.11 or of partial cut-ins by 39% in 11.4. Tesla is amazing how they made it so great before, but could still improve it by this much.

Fantastic.

In fact today, many people are showing videos of their car driving a complete route without an intervention, or at most a bit of encouragement from the accelerator pedal. Sometimes even two drives in a row. That's amazing. OK, it’s true that literally 30 seconds after I got FSD 11.4 it tried to turn left into another car that had the right of way, but it did most of the rest of that trip pretty good. Now, I know that Waymo just announced they are doing 10,000 trips a week, and they don't even have anybody behind the wheel and have probably done 100,000 trips in a row -- but Tesla will get there any day now. As I am about to show, the important thing is that Waymo is doing those 10,000 boring trips a week in just a few cities, while Tesla drivers can have an exciting drive almost everywhere in the USA. Gettin g from 2 trips to 100,000 is no big deal, it's just five little zeros. Six small orders of magnitude. That's nothing for Elon.

Mapping

My Tesla will do FSD in every city, while other cars are geofenced. So while sure, maybe there isn’t actually anywhere my Tesla can drive me without crashing, that’s much better than cars that only do their not-crashing bit inside one of these “fences.” That’s because those cars use detailed maps to understand the road even before they get to it, while Tesla figures it out on the fly from what it can see as it approaches. So while sure, they could let them out of their fence with a supervising driver, they focus on where they have the map. Tesla is trying to solve the hard problem of not needing a map while the other teams focus on the easy problem of making a reliable safe driving system.

And maintaining those maps is just too hard and expensive. These other companies are useless because they are stuck in their fences. Mapping the whole USA is really hard. I mean sure, Google has done that several times but they’re really big and Waymo is not like Google. . MobilEye took as much a s a month to map most of the streets of Europe. That’s like forever. Then sent a team of two to Munich and it was 2 whole weeks before they could drive that town, but a Tesla would drive it with somebody behind the wheel right away. What does MobilEye know? Yeah, they made the original Tesla Autopilot, but then broke off that deal after a driver lost his head, but now they’re only in 50 million cars. And when Cruise expanded to Phoenix and Austin, it took them 3 whole months – more than forever – to do that. OK, Waymo just doubled their Phoenix service area but they still need a driver supervising on the freeways, just like a Tesla. But that’s just a robotaxi, not a car. A taxi service is fine with a map and service area, but that’s not the Tesla way. “There’s going to be a dedicated Robotaxi and it’s going to look quite futuristic.” Sorry, I mean robotaxi is the way to go.

Even so, it’s clear. Driving safely enough to take the human out in a city, that’s the easy part. Scaling up the mapping is the hard part, and that means Tesla will leave these companies in the dust. All Tesla has to do is get from doing 1-2 drives to doing 10,000 and they are golden.

Vision

Tesla’s big advantage is in their AI and their commitment to do it all with cameras and computer vision. Tesla declared back in 2016 that every Tesla sold had the hardware needed for full self-driving and even showed it in a video. OK, sure, they faked that video but the message is clear – cameras are the way to go, even the cameras of 2016. They even took out the radar and the ultrasonic sensors, because as Elon says, “The best part is no part.” The other companies foolishly think digital hardware keeps getting better and cheaper, and so are always upgrading their gear. OK, Tesla has had to upgrade the computers and is about to do that again, as well as the cameras, but Elon has promised the system will still work with the original computer upgrade and the old cameras, and he keeps his promises.

It’s obvious that cameras are the answer. After all, humans drive everywhere with just our eyes, which are just cameras. Well, we also need the human brain, but that’s just a computer. As we know, the biological way is the best way. That’s why all the early aviation pioneers made planes that flap their wings, and why we always make machines that operate as human analogs. All the successful AI thinks about things just the way a human brain does. Well, OK, so there are a lot of “experts” who say that’s actually almost never true, but they’re not the wizards at Tesla.

In fact, they want to use things like LIDAR laser scanners and radars and more, but Tesla has been smart and removed the radar, and never had a LIDAR. That’s because, as Elon says, “LIDAR is a fool’s errand” and a crutch. Sure, LIDAR offers superhuman vision with reliable 3-D depth sensing and it works the same day and night, but it’s not the way humans see the world. Those tempting abilities are a crutch that will distract you from building the AI that can do it all with cameras, Elon says. The last thing you don’t want if your leg isn’t working well is a crutch, after all. If you use a crutch, you’ll never get back on your feet. While computer vision might have a broken leg right now, we know it can work because again … people.

Radar turns out to be a crutch, too. I have a radar in my Tesla but I’m glad Tesla turned it off. Yeah, I get more phantom braking now but the radar was just confusing the cameras with its ability to tell the speed of things, detect hidden vehicles and look through fog. We don’t want superhuman abilities. Humans don’t need them and machines don’t either. It’s strange, because after the radar in a Tesla with Autopilot avoided a serious accident back in the early days, Elon Musk retweeted the video about it, but he knows better now.

LIDAR is also just too expensive. We need our first robotaxis to be cheap. As we know, electronic devices that are expensive in small quantities stay expensive even when you innovate and start making them by the millions. That’s why back when computer drives cost $100,000 a gigabyte, the smart companies did not plan for them to become affordable – that’s just too expensive for any real product. It’s important to save money now. That way, when your Tesla starts hiring itself out as a robotaxi you’ll make even more money. The other companies are foolishly following a “master plan” of starting at higher cost and better quality to get things out and get them adopted, so they can move to lower cost with new technology and volumes. Tesla would never use that as a master plan.

Safe

Today, Tesla Autopilot and FSD need a driver watching them. This is really safe – Tesla publishes statistics every quarter showing Teslas with Autopilot have a crash about every 4-5 million miles while the government reports all cars have one every 600,000 miles. Tesla is clever though, and their number defines a crash as an airbag deployment, but the government number includes any crash the police saw, and airbags only deploy in a small fraction. See how brilliant Tesla is – by picking the right numbers, we can see that Tesla Autopilot is many times safer than ordinary driving, when it’s actually pretty similar.

The leader

Tesla is sure to win. They did it in electric cars and that means they can do it in every other space. They are the obvious AI leader. OK, maybe Waymo, being part of the Google family gets access to the tech that beat all humans in Go, or a chatbot that fooled an employee into thinking it was sentient, or until recently to their employee Geoff Hinton, who invented all this deep learning stuff that Tesla are now the clear leaders in. But what do they really know about AI compared to Tesla?

Tesla also makes its own processors, renting designs from the finest companies of silicon valley. No way they won’t beat Waymo, which gets Google’s world-leading TPU AI accelerators, and MobilEye, which is owned by Intel, the world’s leading processor company, or NVidia, the ruler of public AI processors. Tesla is the smartest user of these technologies and will crush the people who invented them.

We know they are the leader because Elon has promised it. He’s promised that FSD will drive you around more safely than a human can this year, for sure. Elon’s the world’s top genius and his predictions on AI are as sure a bet as investing in Twitter.

So that explains it. I’m Stanley – Tesla Stanley – signing off and returning you to your regular coverage that just doesn’t understand why Tesla is the clear leader.

Add new comment