In addition to Waymo and Tesla, an invisible giant in autonomous driving has emerged

When it comes to autonomous driving, the companies that come to mind include Waymo, Baidu, Uber, Tesla, Cruise, etc.

However, another "invisible giant" that does not often make headlines is also not weak, and it may even eventually come up with truly disruptive technology.

This company is Mobileye, an Israeli startup that Intel acquired in 2017.

Mobileye doesn't have the aura of Elon Musk, nor is it as wealthy as Google.

However, it has something more important on its hands - absolute dominance over today's ADAS market.

In 2019, Mobileye achieved outstanding results: 17.4 million ADAS systems were sold.

This means that 17.4 million cars are equipped with Mobileye's cameras, chips and software.

Mobileye CEO Amnon Shashua

At this year's CES, Mobileye CEO Amnon Shashua proudly announced that Mobileye has gained a huge strategic advantage.

In addition, Shashua also outlined the company's evolution of autonomous driving in the next five years, and reiterated Mobileye's determination to maintain its core position in the industry.

1. "Level 2+"

In the past two years, Cadillac's Super Cruise system has been the ADAS benchmark on the market. It stands out from the crowd mainly because of two major functions.

Super Cruise system equipped with Mobileye EyeQ3 chip

First, the system integrates a camera facing the driver to ensure that the driver does not lose focus.

This seemingly simple feature solves a major problem that has plagued ADAS systems: drivers become overconfident in the technology and let their guard down.

Second, Cadillac "scouted" American highways in advance, collected a total of 130,000 miles of high-precision map data, and updated the map data every three months.

It should be noted that Super Cruise will only be activated on roads that Cadillac has mapped. If you are not driving within the map range, the system cannot be started.

Moreover, the camera will constantly monitor the road conditions to determine whether the current vehicle can use Super Cruise, so its safety is much higher than that of ordinary ADAS systems.

In his speech, Shashua defined the ADAS system used with high-precision maps as Level 2+, and many manufacturers are competing around this concept.

However, in the current Level 2+ market, Mobileye has already occupied 70% of the market share, and its customers include giants such as Nissan, Volkswagen and BMW.

It should be noted that Mobileye is not simply a supplier that sells goods. With the technology at hand, Mobileye has full bargaining power.

One manifestation is that Mobileye can now collect sensor data from Volkswagen, BMW and Nissan vehicles. In addition, Mobileye has persuaded three other companies to open up data collection rights to Mobileye.

With sensor data interfaces from six manufacturers in hand, Mobileye is already extremely powerful.

Shashua claimed that Mobileye can collect 6 million kilometers of sensor data from vehicles every day. It is expected that by the end of this year, 1 million vehicles in the European market alone will provide data for it, and the US market will also reach this number in 2021.

With data in hand, Mobileye can draw high-precision maps.

Today, Mobileye has software that can automatically generate high-precision maps at an average rate of 45 miles (72 kilometers) per hour.

If the current pace continues, high-precision mapping of the European region will be completed in March this year, and high-precision mapping of the United States will be completed later this year.

Don’t underestimate high-precision maps. In the past, they have always been one of the main obstacles to the implementation of autonomous driving technology.

In the past, if you wanted to draw a high-precision map, you had to send people to drive around the streets to scan and draw, and then use manual labor to mark the collected data, which was both time-consuming and laborious.

If Mobileye can crowdsource the entire process and use automation tools to significantly improve efficiency, the data produced will be invaluable.

After inputting this data into high-precision maps, Mobileye can feedback the real-time situation on the road to vehicles on the road.

In other words, Mobileye's partners can easily enter the Level 2+ era.

2. Autonomous driving is possible with cameras alone

Mobileye's position in the ADAS market is now quite solid. Under the huge inertia, Mobileye may be dragged down by its own advantages.

Currently, Mobileye's business model is to sell chips, cameras and related software to automakers.

However, many industry insiders believe that the first batch of fully autonomous vehicles to be put into use will belong to taxi fleets, rather than ordinary passenger cars sold to the public.

Shashua agrees, saying he doesn't want Mobileye to be left behind by its competitors.

Therefore, in addition to Level 2+ products, Mobileye has also developed long-term plans for fully autonomous driving technology.

At CES, Mobileye exhibited a self-driving car.

What’s unique is that it only needs 12 cameras to achieve autonomous driving, and does not use laser radar or millimeter wave radar.

Judging from the demonstration video released by Mobileye, the prototype car traveled safely through the chaotic streets of Jerusalem for 20 minutes, crossing intersections, merging lanes, and making evasive maneuvers.

Seeing this, I’m afraid that the camp of people who follow Elon Musk’s theory that lidar is useless is about to get excited.

However, although Mobileye can achieve autonomous driving by relying on cameras alone, Shashua is not prepared to really cut out lidar and millimeter-wave radar.

On the contrary, this type of vehicle that relies solely on cameras to solve autonomous driving is just part of Mobileye's safe autonomous driving system.

According to the plan, Mobileye plans to build two completely independent autonomous driving systems:

One is based entirely on cameras, the other on radar and lidar.

If both systems can ultimately achieve 10,000 hours without an accident, then the two systems combined can achieve 100 million hours without an accident (10,000x10,000).

By then, Mobileye can confidently declare that its products far surpass human drivers in terms of safety.

When Mobileye proposed this hypothesis two years ago, many people were skeptical, and now it seems that they are still reluctant to change their minds.

Because Mobileye assumes that the failure modes of the two systems are statistically independent, but lidar may not be able to solve the scenarios that would confuse the camera.

Even if Mobileye's assumptions are not true, redundancy is crucial in safety-critical systems.

Camera-based and lidar-based systems will certainly have different failure modes.

Therefore, building two autonomous driving stacks around different sensors and then running them in parallel does provide some safety redundancy, and Mobileye's assumptions are just overly optimistic.

When designing this autonomous driving system with 12 cameras, Mobileye's pursuit of redundancy can be seen everywhere.

For example, to detect objects around the vehicle, Mobileye even included a suite of 6 different algorithms:

The first algorithm is mainly used to identify the wheels and infer the vehicle's position based on this information.

In addition, Mobileye has algorithms specifically designed to identify car doors, as suddenly opening a car door could pose a potential safety issue.

Another algorithm uses "visual lidar (VIDAR)", which compares images from different cameras and then infers the distance between each pixel in the image.

The algorithm then generates a 3D point cloud based on the existing data. Finally, Mobileye uses standard software to decode the point cloud data to identify objects in the scene.

The third algorithm focuses on identifying pixels that correspond to drivable roads. Pixels in the scene that do not belong to the road may be obstacles that the vehicle must pay attention to.

Mobileye hopes to reduce the chances of misidentification by using different image processing methods. Once an object is detected, four independent algorithms locate it in 3D space.

Admittedly, this "remember to check after the test" approach can indeed improve safety redundancy, but how will the system react if different algorithms cannot reach a consensus?

Unfortunately, Shashua didn't provide a detailed explanation of this issue during his CES presentation.

3. The evolution of fully autonomous driving

As mentioned before, Mobileye has no intention of launching a camera-only self-driving car; it is just part of a more complex system.

"Three cobblers are better than Zhuge Liang", the super system of superposition and integration is the result that Mobileye wants.

Although he has to walk on multiple legs, Shashua hopes to walk faster.

Currently, Mobileye plans to deploy self-driving taxi fleets in Tel Aviv, Israel, Paris, France, and Daegu, South Korea in 2022.

In terms of cost, Shashua hopes that the cost of the autonomous driving hardware for the first batch of self-driving taxis can be controlled at US$10,000 to US$15,000.

Three years later, Mobileye aims to reduce the cost of the entire system to less than $5,000.

Shashua's speech sent an important signal that Mobileye has become a leading player in the industry that cannot be ignored.

Mobileye's dominance in the ADAS market has brought it an important strategic advantage, while its solid relationship with automakers has facilitated the deployment of new technologies. Coupled with its massive reserves of sensor data and engineering technology capabilities, Mobileye's future should not be underestimated.