This article comes from WeChat public account “Car Heart” (ID: Auto-Bit), author Ye Fang, Ai Faner authorized to release, please contact the source.

When it comes to Tesla, everyone will not consciously stand in two camps:

One party took the electric car company to the sky; the other side sang, and thought it would eventually be killed by the opponents.

In other words, Tesla will either change the auto industry once and for all, or it will close in the near future.

However, if we let go of the magical existence of finance, competition, and Elon Musk, we can get more unique interpretations from the perspective of rationality, neutrality, and objectivity.

If Tesla’s technical accumulation in electrification may be caught up by the opponent, then it is absolutely unmatched in the data.

Tesla is using this data to build the world’s most advanced and complex neural network.

  Data company Tesla

Is Big Data still “Star of Tomorrow”?

Silicon Valley likes to use some hot words to describe “the next epoch-making product.”

In the case of big data, it is called “new oil.”

This description is very relevant. The data is like the oil buried in the earth’s crust, waiting for people to develop, refine and utilize it to build their own competitive advantage.

However, many yearsToday, the light of the concept of big data has gradually dimmed.

Because from a technical point of view, how to extract truly valuable parts from massive amounts of data is a huge challenge – whether or not the data has its own organizational structure.

  Data company Tesla

In fact, we have fallen into what Gartner calls big data “trough of disillusionment).

This technology has not become a huge unified data platform as envisioned ten years ago.

After the big data whistle, artificial intelligence (AI) and machine learning (ML) became new scent.

However, unlike social media platforms (just want to optimize algorithms to sell ads), everyone is still attentive and doing something meaningful, Tesla is one of them.

Tesla’s true competitiveness lies in the skilled integration and application of data, AI and ML, which converge into Tesla’s neural network:

A super system that combines sensors, data, communications, CPU, peripheral hardware and software.

These key nodes not only work together to process information, but also adapt and learn like people.

The contest has already started

Wall Street analysts believe that the market potential of autonomous driving is trillions.

This is why Tesla, Waymo and other technology giants and traditional manufacturers are battling all the crazy competition.

In fact, when you think about it, you will realize that the concept of autonomous driving has begun to infiltrate into our driving habits many years ago.

Similar to the popularity of fixed-speed cruising, ABS anti-lock braking and other technologies, in fact, human beings gradually transfer control of vehicles step by step.

And Tesla’s Autopilot, probably the most sophisticated driver assistance system available on the market today.

However, upgrading from semi-automatic driving to fully automated driving is not easy.

How can we ensure that this computer on the wheel can keep thinking, making judgments and making the right decisions in this crazy world?

Know that the maturity of a fully automated driving system requires engineers to invest millions of hours. They have to write code, define and refine algorithms and 3D models, and the simulator and test car can’t stop for a while. .

Of course, this is the biography of Tesla’s eyes.The way, and Musk is best at breaking the rules, he led Tesla to a new path.

After 16 years of development, the number of Tesla electric vehicles in the world has exceeded 600,000.

But these factory vehicles are the data harvesters of Tesla.

A human driver’s every move in the car, such as steering, braking or stepping on the throttle, is the data point that Tesla needs.

The collected data will be “fed” by Tesla to its own algorithm, and the upgraded algorithm will “feed back” to Tesla vehicles through OTA.

  Data company Tesla

As of July this year, Tesla has delivered over 643,000 Tesla vehicles with autonomous driving capabilities, of which 528,000 are equipped with the Autopilot Hardware 2.0 system.

Tesla’s self-driving mileage reached 1.56 billion miles, accounting for 10.8% of Tesla’s total mileage (14.4 billion miles).

Compared to the way, Waymo, the most experienced way to drive a road test, has only 15 million miles of experience.

And Tesla is afraid that the accumulated data collection miles under “shadow mode” is already 10 billion miles.

For Tesla’s neural network, this is truly an unbeatable super treasure.

However, this is not the most important difference between Tesla and other companies. The most frightening thing about Tesla is:

  • Tesla’s data comes from the real world;
  • Car owners commute every day and they train Tesla’s AI/ML engine without knowing it.

Musk once said: “When a car learns a new point of knowledge, all Tesla can take control of it.”

Obviously, Tesla has developed a crowdsourced AI/ML training program that looks the best in the world.

Autonomy Day Seeing

I believe many people will be curious: How does Tesla brush out the data and achieve continuous improvement in system performance through data?

The most dry thing in Tesla’s series of events this year is “Autonomy Day.”

In the keynote speech, Stuart Bowers, Vice President of Tesla Engineering (currently resigned) [1] told usTesla’s mental journey.

“Before we started, we tried to understand the world around us,” Bowers said. “The Tesla electric car comes standard with 8 cameras and 12 ultrasonic sensors (radar), as well as inertial measurement devices and GPS bucks.

In addition, the steering wheel and pedal operation that are often forgotten by everyone must also be taken into account.

Bowers also pointed out that these sensors have an “overlapping area” that allows for double confirmation.

In this way, Tesla “has an extremely accurate understanding of what is going on around it.”

Each event, or a human-computer interaction, is recorded and uploaded to Tesla’s database.

Then, these data will be used for 3D simulations for Tesla software engineers to study how to enhance and refine existing algorithms.

The upgraded algorithm will of course be pushed to each Tesla owner via OTA to further enhance the vehicle’s driving experience.

  Data company Tesla

Shadow mode

Of course, software iterations in the car can’t be as casual as a smartphone, after all, it’s about personal safety.

Here, Tesla intelligently used the “shadow mode” to test the modified system in this mode.

Obviously, this is a huge improvement over simple simulators or on-board testing. After all, shadow mode is running in real time and is closely related to the real world.

However, the “thinking and decision making” of the entire vehicle is carried out behind the scenes, so that a continuous feedback loop can be constructed.

In a nutshell, the shadow mode is like a young teenager. He doesn’t have a driver’s license. He often sits in the co-pilot to observe his father’s every move.

“When there is a new algorithm, we want to try it in the first place. In shadow mode you can push it to the team and see how it performs in the real world,” Bowers explained.

In the end, Tesla was able to use machine learning to have more capabilities, and then proceeded to the stage of deployment, which is Tesla’s “early access program.”

At the moment, Tesla is also testing new behavior prediction functions to make it easier for vehicles to predict the next move of pedestrians or bicycles ahead of time.

“We can detect obstacles on the road, and pedestrians are one of the obstacles,” Bowers said.”Vehicles can indeed see pedestrians and bicycles on the road. Tesla’s next-generation automatic braking system will not only stop pedestrians on the road ahead, but will also automatically pass for those who are about to take the road.”

Bowers revealed that this new feature is now running in shadow mode.

In the future, Tesla will definitely push this feature to every owner. However, before this, “experimental” will be carried out on the hard-core owners who signed the early user participation plan.

Another example is a lane change on a highway.

Tesla says it has successfully completed 9 million lane changes in Autopilot mode.

“We have accumulated 100,000 successful lane change cases almost every day.” Bowers said.

In Bowers’ view, the real end of the battle is “to integrate neural networks, vehicles and all data to create the ultimate truth for helping vehicles understand the world.”

  Data company Tesla

Mobility as a Service

The advent of autonomous driving means that the original vehicle sales model will collapse completely and replace it with a new era of everyone’s taxiing, which is what we call travel-as-a-service (MaaS).

In a recent interview, Tasha Keeny, an analyst at ARK, an investment company focused on innovative technology and markets, pointed out:

In a sense, MaaS is already on the road. After all, Uber has already reached a milestone of providing 1 billion travel services in a single quarter, and Uber-like taxi service is very popular around the world.

However, Keyny’s data shows that “renting” a car is still more expensive than buying a car.

If you buy a car, the average cost per mile is 70 cents, which is cheaper than going out to take a taxi.

Of course, this phenomenon will be completely reversed after the automatic driving, and the MaaS will be 22 cents per mile after the human driver is removed.

And don’t forget that the millennial generation, which is in the prime of life, is also familiar with the sharing economy.

Keeny believes that behind the turning point, the use of mobile phones to call auto-driving cars will create a supermarket with more than $5 trillion.

This is also the main reason why manufacturers are competing for autopilot highlands – otherwise, which investors can suffer Uber’s single-quarter loss of $5 billion.

Tesla will of course participate in the grand drama of MaaS:

Tesla wants to “requisition” the owner’s car to deploy an autonomous fleet, with the goal of a win-win situation.

The owner’s vehicles do not have to park in the parking lot for a long time. These vehicles can go out and work while the owner is working, and Tesla can earn a service fee.

The bet is really high, but the loot is equally rich.

Once the world of MaaS is officially completed, the rules of the game will change radically. From a technical point of view, Tesla does have a huge lead.

The current traditional car manufacturers have done their best to design and build the best vehicles, while cutting costs through mass production to gain a competitive advantage.

This is also their best choice for the moment.

However, industry spoilers are not bound by these established rules, and Tesla is such a company that does not take the usual path.

Self-developed chips, hardware, software, and their own neural network and MaaS fleet, Tesla has firmly grasped the fate in his own hands, and the role of data is to thread the needle, it is to drive the entire system operation The aorta.

As for the capacity issue, it is only the one with the lowest weight in the whole process.

If you still don’t believe it, look at the Shanghai Super Factory.