Published on April 23rd, 2019 | by Paul Fosse0
Can Tesla Pull Ahead Of The Industry And Deliver Full Self-Driving Cars Next Year?
April 23rd, 2019 by Paul Fosse
In this article, I will examine the claims that Tesla made at its Tesla Autonomy Day on April 22. Although I’m a small investor in Tesla and undoubtedly a fan of the company and its cars, I’ll try to be as objective as I can be and show where Tesla’s claims are proven or undisputed and where they are unproven and require a leap of faith. In order to have a robotaxi, Tesla has to have the following pieces of the puzzle:
1. Cars. Tesla claims to have built about 500,000 vehicles that have the necessary sensor array (any Tesla built since October 2016 has 8 cameras, 12 ultrasonic sensors, and 1 radar), and these cars can get the upgraded Full Self Driving (FSD) computer. Tesla has plans to make another 500,000 cars in the next year, so it should have about a million cars available to compete with Uber and Lyft on the flick of a switch.
Of course, Uber and Lyft have more than a million cars signed up on their platforms (thought to be over 2 million just on Uber). Not all of those million Tesla cars are willing to let a stranger ride in them for money. On the other hand, Uber and Lyft drivers only drive a few hours a day, whereas the Teslas can run up to 24 hours a day, since they won’t need a driver.
It seems clear that Tesla won’t have the scale to do significant damage to Uber and/or Lyft when it first rolls out services. On the other hand, investors attempt to anticipate the future with their investments, and if they perceive Tesla’s story to be credible, it will do great damage to Uber and Lyft — if Tesla can scale over the next 5 years without paying a cent to drivers, it is obvious the company will have much lower costs than Uber and Lyft, unless they can find access to millions of self-driving cars.
2. Redundancy. Tesla needs cars that can accelerate, brake, and steer using electric motors. Elon and team mentioned that they have full redundancy in braking and steering (they didn’t mention acceleration), so that they can have a failure in a power steering motor and a power braking motor and still safely steer and stop. I would think you would want to then stop and have the problem fixed, not continue with the single steering and brake motor. Although this can be done with gas cars, most people claim control of a car is slightly easier with electric cars. This point isn’t really disputed by the company’s critics.
3. Electric cars versus gas or diesel cars. You can certainly build a self-driving gas or diesel car (if you can figure out self-driving of course), but it is undisputed that electric cars have much lower fuel costs (about a quarter the costs in most areas). If you only drive a few miles a day, that tends to counterbalance the higher initial purchase price of the electric car. If you drive the car a lot, as in 24 hours a day to maximize income, the lower costs of an electric car become very significant. Tesla is the only electric car manufacturer in the US that has significant scale. It appears automakers across the whole industry are electrifying their lineups, but it is widely disputed how quickly this will occur, and even if it will occur.
Elon made a claim that their Model 3 motor and body can go a million miles and that their battery packs can go 300,000 to 500,000, but that is unproven. At the presentation, Elon claimed a new battery pack was coming out next year designed for more charging cycles so that it would last a million miles. This is unproven, but Elon’s record with these sort of claims is excellent. He has always delivered the promised battery performance, but not always in the promised timeframe. It is thought that maintenance costs on Tesla vehicles are much lower than gas cars and although that advantage is sometimes disputed, the evidence is quite strong that it exists.
4. Sensor array. Does Tesla have the right sensors?
Nobody disputes that cameras, ultrasonic sensors, and radar are very useful, but almost everyone thinks lidar is needed. I’ve written about it here. CleanTechnica has also addressed it here and here. The issue is that although lidar makes it easier to find the safe areas to drive, since it give you a 3D map of the space without using any artificial intelligence (it just shines a laser and measures the time for it to bounce back), it doesn’t work in poor weather and it doesn’t help with many other problems you need to solve to do self driving. The lasers don’t help with stop signs, or traffic lights or recognizing bicycles or pedestrians or cars or predicting future behavior of any of those three. Lidar doesn’t help read road markings or signs or any of the aids used worldwide to aid the billions of human drivers.
Lidar is great if you just want to put a car in a science project and have it wander around on the road and not run into any stationary objects in perfect weather. Then you don’t need any fancy software, you can just tell the car where the stationary objects are and figure a path around them.
As you can plainly see (pun intended), lidar doesn’t help a bit with all of the problems dealing with complex urban environments — namely, moving people, bikes, animals, cars, and trucks controlled by people or animals that do unpredictable things in all kinds of weather. For that, you are going to need some intelligence, either human or artificial.
— Anner J. Bonilla🇵🇷🛩️🔋🔧 (@annerajb) April 23, 2019
That’s Anthony Levandowski speaking above.
5. Intelligence to understand the environment around the car. For you to understand the road ahead, Tesla claims you need some modest CPU and graphics processing power and a massive amount of multiplication and addition power for linear algebra. As I wrote, almost a year ago, Tesla looked at what the industry had available to meet its computing requirements and didn’t find anyone working on a chip that met its performance requirements (especially the ability to process a single image at a time — instead of batches of 256 images — at very low power). If you use too much power, you impact the driving range of the car significantly.
Elon recruited a top team with experience from Digital Equipment, Intel, Apple, and AMD to make a custom chip. Since they had modest requirements for CPU and graphics needs, they licensed existing designs and just put them on their chip. But since they had unique needs for high performance multiplication and addition operations at very low power and they couldn’t find any acceptable solutions available to license, they designed a very simple processor with very high performance. It is a well known truism in the semiconductor industry that you can make a chip faster for one operation if you don’t need to handle a complex instruction set.
You also see this in cryptocurrency mining. If you are willing to design a chip to do the mining, you can perform the operations much faster and using less power than using CPUs or GPUs to do the math operations. The reason every currency doesn’t have a chip designed to mine it is that it costs a fair amount to design each chip and it is difficult to predict which cryptocurrencies will be used enough to pay back that initial chip design cost. Of course that isn’t a concern here — if this chip solves the full self-driving problem, nobody disputes there is tremendous demand.
I’ve heard critics claim that it is unlikely that Tesla could design a chip that is better than the “experts” at Intel and AMD, but I find their plan viable for several reasons:
- They hired top notch talent from the industry.
- They only custom designed the parts of the chip for which they had unique requirements. They licensed proven (but not leading-edge) designs for the CPU and graphics engines. This project would have been much riskier if they had custom designed the whole chip.
- They manufacture the chip at a Samsung fab. Elon may love vertical integration, but he is smart enough to realize that building a 14 nano-meter lithography process moving to a 10 nano-meter process is a headache they didn’t need to take on.
6. You will need lots of training data.
There is no doubt that Tesla has many more cars driving around with cameras than all other players in the world combined. It is disputed whether they can afford the cell phone data charges to send the data all back to the mother ship. If they can’t (and it is likely they can only send a small fraction of the data back), are they selecting the right sample to get the edge cases they need to make the cars safe? They use driver disengagements to help them decide what is routine video that they don’t need and what is something special that they need to look at and train the image recognition software to deal with.
7. Image recognition and depth perception.
Elon made it clear in the presentation that Andrej Karpathy wasn’t just a PhD and professor of artificial intelligence teaching at Stanford, but that he developed the very popular class taught there on image recognition and is arguably the top expert in the world on training neural networks to recognize images. I think few would dispute that Andrej is a top expert, but many (including me) are unconvinced that image recognition will be able to advance as quickly as Elon claims. I’ve read many articles on this and it all sounds plausible, but it is such a big leap in capability that I can’t help but doubt if they can make this much progress in such a short period of time.
I will say that an example where my skepticism was misguided was the natural language capability of Alexa. I had seen 30 years of PC products that claimed to do speech recognition and they all took a lot of training for disappointing results. Then, all of a sudden, Alexa (and I’ve heard Google has a good one too) solved the problem and it seems to understand what I’m saying pretty well. It still seems pretty dumb at doing complex tasks, but it does a great job on simple ones.
Tesla has a great team, but this problem is just incredibly difficult. This is really the area Tesla just has to prove it works because the world isn’t going to trust them, no matter what they say.
8. Driving the car once you recognize what objects are out there and where they are going. This isn’t too hard with the exception with the game of chicken that drivers play in trying to change lanes. Tesla will have to prove they can find a way to be assertive enough to merge into a crowded lane without causing a minor accident. This is hard for humans and it will be hard for computers too.
I came away from the Tesla Autonomy Day impressed with Tesla’s strategy and enthusiasm, but unconvinced that they will be able to pull it off in the next year. In my 35 year career in software development, I’ve seen many examples of a project that I would expect to take 4 years, be completed in a year with excellent leadership and programming talent. I’ve also seen several projects that could have been completed in a year be cancelled after several years unfinished, usually due to leadership that had great vision but insufficient talent to pull it off. Overly complex development processes have also killed some projects, but I don’t expect that to be a problem at Tesla. Elon has been developing commercial software since he was 12 — he won’t let a bad process kill this project.
My opinion is they can pull it off, but I really don’t know if they can do it next year or not. As Yogi Berra said, “Predictions are difficult, especially about the future.”
If you want to take advantage of my Tesla referral link to get 1,000 miles of free Supercharging on a Tesla Model S, Model X, or Model 3, here’s the link: https://ts.la/paul92237 (if someone else helped you, please use their code instead of mine). I encourage you to buy before the price of Full Self Driving (FSD) goes up on May 1st if you believe in Tesla’s ability to get it working soon.
Follow CleanTechnica on Google News.
It will make you happy & help you live in peace for the rest of your life.