Can Tesla get out of the industry and deliver self-driving cars to the world next year?

April 23, 2019 by Paul Fosse

---

In this article, I will review Tesla's claims during his Tesla Independence Day on April 22. Although I am a small investor in Tesla and undoubtedly a fan of the company and its cars, I will try to be as objective as possible. I can be and show where Tesla's affirmations are proven or not, where they are not proven and require an act of faith. To have a robotaxi, Tesla must have the following pieces of the puzzle:

1. Cars. Tesla claims to have built about 500,000 vehicles with the necessary sensor network (each Tesla built since October 2016 has 8 cameras, 12 ultrasonic sensors and a radar), and these cars can be equipped with the best computer Full Self Driving (FSD). Tesla plans to manufacture an additional 500,000 cars over the next year. It should therefore have about a million cars to compete with Uber and Lyft in a jiffy.

Of course, Uber and Lyft have more than a million cars registered on their platforms (over 2 million in Uber). All these millions of Tesla cars are not willing to let a stranger drive them for money. On the other hand, the Uber and Lyft drivers only drive a few hours a day, while the Teslas can ride up to 24 hours a day because they will not need a driver.

It seems clear that Tesla will not have the scale needed to cause significant damage to Uber and / or Lyft when it first launches its services. On the other hand, investors are trying to anticipate the future with their investments. If they believe that Tesla's story is credible, it will hurt Uber and Lyft greatly – if Tesla can evolve over the next five years without paying a cent to drivers. It is clear that the company's costs will be much lower than those of Uber and Lyft unless they can access millions of autonomous cars.

2. Redundancy. Tesla needs cars capable of accelerating, braking and driving with the help of electric motors. Elon and his team reported that they had full redundancy in braking and steering (they did not mention acceleration), so they could experience power steering and engine failures. assisted braking while maintaining a safe steering and stopping. I think you would then want to stop and fix the problem without continuing with the unique steering and brake motor. Although this can be done with gasoline cars, most people claim that controlling a car is slightly easier with electric cars. This is not really disputed by the critics of the company.

3. Electric cars versus gasoline or diesel cars. While you can build a gasoline or diesel car (if you can handle it, of course), there is no question that fuel costs for electric cars are much lower (about a quarter of the costs in most areas) . If you drive only a few kilometers a day, this tends to offset the higher initial purchase price of the electric car. If you drive a lot, 24 hours a day to maximize your income, the lower costs of an electric car become very important. Tesla is the only manufacturer of electric cars in the United States to have a significant scale. It seems that automakers from across the industry are electrifying their queues, but the speed with which this is going to happen is disputed, even if it will happen.

Elon claimed that his engine and model 3 body could travel millions of miles and that their batteries could go from 300,000 to 500,000, but this has not been proven. During the presentation, Elon claimed that a new battery pack, intended for multiple charge cycles and intended to last a million kilometers, would be released next year. This has not been proven, but Elon's record with this type of claim is excellent. He has always delivered the promised battery performance, but not always on time. It is thought that the maintenance costs of Tesla vehicles are much lower than those of petrol cars and, although this benefit is sometimes disputed, there is ample evidence that it exists.

4. Network of sensors. Does Tesla have the right sensors?

Nobody disputes the fact that cameras, ultrasonic sensors and radar are very useful, but almost everyone thinks that lidar is necessary. I wrote about it here. CleanTechnica also touched on here and here. The problem is that, although lidar makes it easy to find safe areas to drive, because it gives you a 3D map of space without using artificial intelligence (it just shines a laser and measures the time it takes to do it bounce). does not work in bad weather and does not solve the problem of many other problems that you must solve to drive autonomously. Lasers do not help stop signs, traffic lights, recognition of bicycles, pedestrians or cars, nor predict the future behavior of any of these three models. Lidar does not help to read markings, signs or aids used around the world to help billions of drivers.

Lidar is ideal if you simply want to put a car in a science project and let it walk on the road without falling on immobile objects in perfect weather. In this case, you do not need sophisticated software, you can simply tell the car where the stationary objects are and trace a path around them.

As you can clearly see (pun intended), lidar does not solve the problem of complex urban environments: moving people, bicycles, animals, cars and trucks controlled by people or animals doing things unpredictable. in all kinds of weather. For this you will need intelligence, human or artificial.

And now?
What a coincidence it was made live today …
So we have pronto, zoox, commaai and tesla …
#TeslaAutonomyDay pic.twitter.com/9B8GLkYONL

– Anner J. Bonilla???️?? (@annerajb) April 23, 2019

It's Anthony Levandowski who speaks above.

5. Intelligence to understand the environment around the car. To help you understand the road, Tesla says you need modest graphics processing and processor power, as well as a considerable amount of multiplication and addition power for linear algebra. As I wrote, almost a year ago, Tesla looked at the resources available on the market to meet its computing needs and found no one working on a chip that met its performance requirements ( in particular the ability to process a single image at a time – instead of batches of 256 images – at very low consumption). If you use too much power, you greatly reduce the autonomy.

Elon has recruited a high level team with experience at Digital Equipment, Intel, Apple and AMD to create a custom chip. Because they had modest CPU and graphics needs, they licensed existing designs and simply inserted them into their chips. But because they had unique needs for high-performance multiplication and very low-power additions, and they could not find acceptable solutions under license, they designed a very simple, high-performance processor. It is well known in the semiconductor industry that you can build a chip faster for an operation if you do not need to handle a complex set of instructions.

You also see it in cryptocurrency exploration. If you are ready to design a chip to perform extraction, you can perform the operations much faster and with less energy than using processors or GPUs to perform mathematical operations. The reason why each currency has no microchip to design is that the design of each chip is quite expensive and it is difficult to predict which cryptographic currencies will be used enough to pay the initial cost of chip design. Of course, this is not a problem here: if this chip solves the problem of autonomous driving, no one disputes the fact that there is a huge demand.

I've heard critics say that it was unlikely that Tesla would design a chip better than the Intel and AMD "experts", but I find their project viable for several reasons:

• They hired unparalleled talents from the industry.
• They only customized the parts of the chip for which they had unique requirements. They have licensed proven (but not state-of-the-art) designs for CPU and graphics processors. This project would have been much riskier if they had designed the whole chip.
• They manufacture the chip in a Samsung factory. Elon may like vertical integration, but he's smart enough to understand that building a 14-nanometer lithography process to a 10-nanometer process is a puzzle that they did not have to face.

6. You will need a lot of workout data.

There is no doubt that Tesla has many more cars with cameras than any other player in the world combined. The question of whether they can afford to pay the cell phone data charges to return all data to the mother ship is disputed. If they can not (and it is likely that they can only return a small fraction of the data), do they select the right sample to obtain the extreme cases necessary for the safety of cars? They use the disengagement of the pilots to help them choose the routine videos they do not need and what they need to watch and train the image recognition software.

7. Image recognition and depth perception.

Elon clearly explained in his presentation that Andrej Karpathy was not just a PhD student and professor of artificial intelligence at Stanford, but that he had developed the course very popular taught in this field on image recognition and that he was arguably the world's leading expert in networking to recognize images. I think that few people will dispute the fact that Andrej is a leading expert, but many (including myself) are not convinced that image recognition will move as fast as Elon claims. I have read many articles about this and it all seems plausible, but it's such a big leap in capabilities that I can not help but doubt if they can make as much progress in so little of time.

I will say that an example where my skepticism was wrong was Alexa's natural language ability. I had seen 30 years of PC products claiming to do voice recognition and they all took a lot of training for disappointing results. Then suddenly, Alexa (and I heard that Google had a good one too) solved the problem and he seems to understand pretty well what I'm saying. He always seems stupid enough to do complex tasks, but it does a good job with simple tasks.

Tesla has a great team, but this problem is incredibly difficult. It's really the region Tesla just needs to prove that it works because the world is not going to trust them, no matter what they say.

8. Drive the car once you have identified what objects are there and where they are going. It's not too difficult, with the exception of the hen game that drivers play to try to change lanes. Tesla will have to prove that he can find a way to assert himself enough to blend into a crowded lane without causing a minor accident. It's hard for humans and it will be difficult for computers as well.

Conclusion

I came back from Tesla's Independence Day, impressed by Tesla's strategy and enthusiasm, but I'm not sure they will be able to succeed next year. Over the course of my 35-year career in software development, I've seen many examples of a project that I believe will last four years and end in one year with great leadership and leadership talent. programming. I have also seen several projects that could have been completed in one year be canceled after several unfinished years, usually because of a leadership with a great vision, but a lack of talent to make it happen. Overly complex development processes have also killed some projects, but I do not expect it to be a problem for Tesla. Elon has been developing commercial software since the age of 12 – it will not let a bad process kill this project.

My opinion is that they can do it, but I really do not know if they can do it next year or not. As Yogi Berra says, "The forecasts are difficult, especially with regard to 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 anyone any other helped you, please use their code instead of mine). I encourage you to buy before the price of Full Self Driving (FSD) increases on May 1st if you believe in Tesla's ability to make it work quickly.

Keywords: Andrej Karpathy, Elon Musk, Pete Bannon, Tesla, Tesla Autonomy Day, Autopilot, Tesla Full Self-Driving