This week Tesla organized its "Independence Day" at its headquarters in Palo Alto, to give investors more details about the company's automatic steering technology (AutoPilot).
Elon Musk, and the chiefs of the major divisions, were able to deliver nearly four hours of lecture, fashioned from Apple's lectures, and far from the world's standards of the automobile. A conference that was announced as regular, like the F8 for Facebook or Google I / O.
Tesla managers have raised the veil over what experts have already imagined The tremendous leadership of Tesla in the development of its software and equipment of autonomous direction.
At one time or several manufacturers, including PSA, announced at the last Geneva Motor Show, reduce their plans for autonomous cars to reach level 5, the last level of autonomy, do not leave the path completely clean Tesla?
It's true that Tesla's approach, begun more than 3 years ago, ends today in a computer (FSD for Full Self-Driving), which has been integrated in all vehicles since April 12, and will allow a completely autonomous driving through future software updates automatically submitted by Tesla. This platform strategy puts pressure on your competitors. Every day that passes Tesla sells "potential" standalone cars and will have to decide an upgrade, a priori free and more. But this strategy is also pressing the legislator in every state of the United States, which in 2020 will be considered by the American opinion as one of those who will block the process when Tesla says it is ready …
Tesla recalled the goals they had set and which they all espoused. The update is expected in 2019 and the case of the uses of "robotaxis", which reached the headlines for 2020.
Recalling Apple's strategy, Tesla's strategy appears to want to master an integrated platform, including hardware (its Model3 car and FSD card), Autopilot software and the remote car update network. Only applications are missing. The creation of the "Tesla Network," the network of cars your customers can rely on to create the world's largest taxpayer fleet, will be one of those large-scale autopilot applications.
But these robots divert us from the essential. What other manufacturers do not have is simply "the Tesla chip" and its architecture.
This architecture is a break from Lidar's approach (the little bump on the top of the car, the Waymo), which seeks to shape the world and move in. This is the approach currently used by Google. Tesla is looking for an architecture to process much more information per second.
Elon Musk puts 100% in artificial intelligence with an already very operational field, image recognition and non-modeling. It allows you to react to unforeseen events that can be recognized, but also to recognize future situations that are likely.
Tesla therefore placed the AI together with the real-time processing power of the on-board computer.
The Model S / X computer comes from the partnership Mobileye (bought back $ 15 billion Intel in 2017). But Tesla's vision went well beyond that and especially quicker. The company therefore embarked on the development of its own system after recruiting Pete Bannon (formerly Apple and former Intel) who presented himself the results of his team for 3 years. Mobileye, of course, is still in the race, but is just looking for a commercialization of its next chip in 2022 and was left behind by Tesla.
To take into account the criticality of real-time imaging that will drive the car, Elon Musk has not forgotten that he also sent rockets into space and developed a redundant architecture with two neural processors, a dozen CPUs and GPUs adapted to image processing, which constantly compare their results (security system). The data is also encrypted.
At your next town's dinner with geeks, you might ask the question, "Who makes the most powerful processor board in the world?" We should answer you nVidiabut proudly you will be able to say it FSD is almost seven times faster than nVidia's fastest card (144 versus 21 TFlops). There you understand how high the bar is to reach level 5 for standalone cars and maybe why other builders looking for a supplier, sees this with a distant horizon.
Now, with such power, you will need food, you also understand that the second masterpiece is still the drums, which Tesla learned to master!
Andrej Karpathy took over to explain the neural part of the system, programmed to detect objects and situations to make predictions and choose trajectories to send to autopilot. This is where Tesla has another advantage over its competitors with a single platform they dominate and they have access to data (Mobileye is a subcontractor and does not access data from its manufacturing customers, and nVidia is even more advanced). The more than 100 million kilometers traveled on autopilot will be exploited by the neural part to improve its predictions.
If you now agree that full-time driving permission will be somewhere in the world, Tesla can offer its Model3 customers with Autopilot, use their cars to survive, serving as a taxi. These are the famous taxirobot advertised and "Tesla Network" the taxi platform that will be proposed in 2020 and that will certainly be the subject of Tesla next day next year. According to shared calculations, Model3 owners could generate up to $ 30,000 per year for a $ 35,000-bought car that would operate 16 hours a day, enough to allow them to invest instead of Tesla to create the "Network Tesla ".
Of course, this is being presented by Tesla to drive sales, as competitors in the electric sector, such as GM's eTron, will not generate those revenues if you believe them. But it is especially a sacred pavement in theUberbecause it directly attacks its business model of organizing a fleet of taxis that we do not have. The latter, which intends to enter the stock market, has just set its valuation at $ 80 billion, while last year the numbers of 120, and up to 100 months ago, circulated.
What GreenSI retains the underlying potential flaws, in addition to the FSD that makes Tesla the constructor of the most powerful neural map, is that the car comes to take the place of the conquest of space to advance the computers. Software Appolo, to go to the moon, had in its day driven computers, its computing power and especially its miniaturization in terms of sausages. Thanks to them, Ms. Michu could have had a calculator in her living room since 1972. Then follow the progress of the autonomous car because In a few years, they will certainly also drive innovation in other industries.
The second lesson is more forward looking. Elon Musk could create a "Tesla Computer Vision" division, separate from Tesla and assume the FSD and part of the core infrastructure. It would be more valued by investors. And so, how? AWS which today brings back the margin ofAmazonasperhaps this division brings the margin back to Tesla as electric cars invaded the planet and the price war continues. And then, like the debut of Amazon, Tesla's car production model is not yet profitable …