Intel's self-driving tech division Mobileye is picking up speed, set to become a true...

[Pete Flint]

Highly anticipated: Tesla has long overtaken the headlines around self-driving vehicles, but Intel's autonomous vehicle division Mobileye is laying the groundwork to be its biggest competitor. Mobileye's CEO revealed its upcoming lidar tech at CES 2021, and marked an exciting, differentiating strategy between the industry's leading giants.

The world of self-driving vehicles is roughly hewn into two factions: companies that believe fully autonomous vehicles will naturally evolve from advanced driver-assistance systems (ADAS), and those that expect the two will exist in concert, fulfilling different niches.

The latter is currently led by Alphabet’s self-driving outfit Waymo, which has a testing ground of self-driving taxis in the suburbs of Phoenix as well as scattered tests within San Francisco. The former however is where companies like Tesla reside, as well as their main, lesser-known competitor in this space, an Intel division called Mobileye.

These two exist in a hypothetical space where they race for dominance over the inevitable, self-driven future of automotive travel. While it is possible that they are scrying the future of this industry correctly, their tactics and investments are quite different.

Israeli tech firm Mobileye was acquired by Intel in 2017 for $15 billion and has become the largest supplier of ADAS systems installed in modern cars. Mobileye had a big CES showcase this year to explain how they believe their strategy for comprehensive self-driving vehicles will not hit the “glass ceiling” that awaits Tesla, according to Mobileye’s CEO, Amnon Shashua.

Creating a system for self-driving vehicles requires a large foundation of data to train a company’s proprietary AI which will allow a vehicle to recognize and react to variation on the road faster than a human is capable of. Shashua and Tesla’s Elon Musk approach this challenge differently.

Tesla’s datasets comprise countless hours of video information recorded by its fleet of cars, which the vehicles selectively send to Tesla engineers when Wi-Fi is readily available. Tesla’s team can then query this information and extract clips which are the most optimized for training its platform. An alternative to this video-only strategy is to use more expensive lidar sensors. Back in 2019 however Musk was quoted saying, “anyone relying on lidar is doomed.”

Mobileye’s Shashua believes that a more practical, reliable system would utilize both. Their ubiquitous ADAS systems do not currently use lidar, but in an interview this week the CEO showed off a new lidar-based system, which would be implemented into their vehicles by 2025. Mobileye used to be a Tesla partner up until 2016, but today it's working with several automakers including BMW, Ford, Nissan, and Volkswagen.

The Intel subsidiary does not believe they need to use the “brute force” strategy of AI video-training that Tesla implements, but rather the “semantics of the road” can be discerned through more targeted means. Their current systems record video data, which it processes locally to understand the geometry of its environment, then sends those small packets of information back to Mobileye to help construct massive 3D maps. These maps provide detailed geometrical data as well as contextual, behavioral data in local regions of the city or street they were taken.

Mobileye does not expect to combine its lidar and video technology until both are ready for prime time, at which point it aims to provide one comprehensive system for self-driving vehicles. Only time will tell if their strategy to partner with large automotive companies and develop maps of cities around the world will outcompete the brute force of Tesla’s video-based system on its own vehicles.

If lidar proves essential in future, then Tesla might see difficulties ahead. Waymo, too, may find trouble expanding their solution if their technology is too specialized to the regions in which they test. Mobileye, a firm with lesser name recognition than the former two, may have a platform that is comprehensive and global enough to outlast or even surpass its competition.

Permalink to story.

https://www.techspot.com/news/88312-intel-self-driving-tech-division-mobileye-picking-up.html

 

[Uncle Al]

Looks promising. Ideally, at some point in time a "standard" will be developed with shared technology, data, etc. so these systems will be similar (sort of like seatbelts). What will be really interesting is if cars can be set up with all the necessary connection points so a unit can be bought, plugged in and used, as needed at some future date in case the driver cannot afford it in the inital price. Of course we are probably talking 30-50 years down the road but, if encouraged correctly, it would be a huge benefit to the public and provide a financial incentive to the dealers .....

 

[VitalyT]

We have decades of experience in computer vision

That's where they lost me. Was their first prototypes based on 35mm film analysis?

Absurd title, absurd claims, no real product, just self-advertising click-bait.

 

[Plutoisaplanet]

MobilEye is the only other company with a good strategy towards solving self-driving besides Tesla. They are approaching this as a vision-first solution, which is exactly how roads work and the best way to distinguish objects (ie. tumbleweed vs. plastic bag). LiDAR is only helpful in very accurately mapping out everything on the road, but not to identify everything. As proof of this, see if you can effectively drive with blurry vision/look at the road with your eyes out of focus (hint: you can still drive perfectly).

 

[Tom Yum]

MobilEye is the only other company with a good strategy towards solving self-driving besides Tesla. They are approaching this as a vision-first solution, which is exactly how roads work and the best way to distinguish objects (ie. tumbleweed vs. plastic bag). LiDAR is only helpful in very accurately mapping out everything on the road, but not to identify everything. As proof of this, see if you can effectively drive with blurry vision/look at the road with your eyes out of focus (hint: you can still drive perfectly).

This assumes an either/or approach that no-one except Tesla is advocating for. Lidar only has flaws, vision-only has flaws, but there is no reason they can't be complementary. Tesla's issues with running into white trucks for example is an issue with visual-only systems, it couldn't tell the difference between the white truck and the white clouded sky above it. Now we could argue a human wouldn't make that mistake, but then again the human eye provides a contrast ratio of over 1 million to 1, compared to the latest camera tech which struggles to get 5000 to 1. Lidar, radar and visual sensors all augmented together would provide a sensor system that exceeds human capability, which is important. The end game of automated driving is not just to be better than humans, but as safe as possible. It is the model Waymo uses, and probably goes some way to explain why their tech is currently the most successful of all the self-driving systems.

The other thing that is strange about this article is that there isn't a single mention that Tesla used Mobileye tech in its first iteration of auto pilot, but moved away from their tech for its second iteration. Odd not to at least mention that when talking about them as being competitors.

 

[Sedziwoj]

Odd not to at least mention that when talking about them as being competitors.

they mention it

Mobileye used to be a Tesla partner up until 2016

Problem with Mobileye, and articles as this is, that they don't read between lines. As Mobileye for full autonomy will use lidar and cameras, but from article it get on roads (and I bet it mean, as assistance system, not fully working FSD) in 2025, and they only then start to collecting data.
They are using hand make code, which for many years is accepted that it have "glass ceiling", AI (DNN) required more data and processing, but is closer to what humans do, so less dangerous of hitting "glass ceiling" (Tesla should add one/two more thing to get all things humans do while driving)

 

[seeprime]

That's where they lost me. Was their first prototypes based on 35mm film analysis?

Absurd title, absurd claims, no real product, just self-advertising click-bait.

It's PR designed to keep restore faith in Intel's stock prices, which dropped 20% in 2020. Nothing more at this time.

 

[yeeeeman]

Having map is the best way to tackle this. Sure, you have your own algo of handling new situations, but a map greatly simplifies things.

 

[mosu]

The real challenge would be testing the systems in India or Turkey, not in San Francisco. But maybe they do not have enough cars...

 

[Tomfoolery]

That's where they lost me. Was their first prototypes based on 35mm film analysis?

Absurd title, absurd claims, no real product, just self-advertising click-bait.

Maybe do some research before posting stupid comments. Mobileye was formed in 1999, you know like over 2 decades ago.
Mobileyes ADAS systems are in more cars than any other system.and is the basis that Tesla autopilot was built on. If you watch any of their videos your I'll see that their system is way more advanced than FSD is. Even recognises actions that pedestrian and other driver are doing based on their body language.

 

[Tomfoolery]

they mention it

Problem with Mobileye, and articles as this is, that they don't read between lines. As Mobileye for full autonomy will use lidar and cameras, but from article it get on roads (and I bet it mean, as assistance system, not fully working FSD) in 2025, and they only then start to collecting data.
They are using hand make code, which for many years is accepted that it have "glass ceiling", AI (DNN) required more data and processing, but is closer to what humans do, so less dangerous of hitting "glass ceiling" (Tesla should add one/two more thing to get all things humans do while driving)

Mobileye has full autonomy now with a camera only solution, but if you read their presentation you would know that they want to add lidar on order to achieve full sensor redundancy. As not being as stubborn as Elon, who has painted himself into a corner, they have realised that regulators are likely to require cameras, radar and lidar for redundancy. Especially as every other AV is using lidar. All Tesla's development is going to turn out to be useless and all the people who bought cars expecting autonomy are going to very disappointed.

 

[VitalyT]

Maybe do some research before posting stupid comments.

Maybe read some forum rules before posting things like that. Your first post on TechSpot, and right off the bat it is a personal attack, which is very much frowned upon here, because it's just rude.

 

[Sedziwoj]

Mobileye has full autonomy now with a camera only solution, but if you read their presentation you would know that they want to add lidar on order to achieve full sensor redundancy.

In DNN you can't easily take important system out, or they will have bigger error rate = not safe. And for redundancy, you adding 2nd pair of sensors and connection and processing, not different sensors. They said it this way, so people who have not idea how it works, think they are good.
Think, why add Lidar which cost few thousand dollars, when you may add second camera for $5?
PS "Elon" is using Lidar own construction, but is for Crew Dragon.

 

[Markoni35]

They put a little person in the trunk, which then drives with the help of GPS and cameras. That's why it's got those holes... for air.