A Tesla-Beating Self-Driving Leader Just Made A Catastrophic Mistake
A Tesla-Beating Self-Driving Leader Just Made A Catastrophic Mistake
Or have they?…
You have most likely never heard of Mobileye. However, they pioneered self-driving technology years before the Tesla Model S came out. You have never heard of them because they sell their technology to manufacturers like BMW, Audi, Volkswagen, Nissan, Ford, Honda, General Motors and more; as such, over 170 million vehicles worldwide use Mobileye automated systems. Several already operating robotaxi services run solely on Mobileye sensors, chips and AI. Meanwhile, Tesla has only sold 5.8 million cars in total and has no robotaxis. This gives Mobileye a vast leg-up over Musk. In fact, their EyeQ6 and EyeQ7 systems are set to embarrass their US counterpart, as they are predicted to be one of the first 100% fully automated self-driving systems on the planet. But they may have just made a catastrophic mistake that puts this promising future out of reach.
What’s this decision? Well, Mobileye has decided to disband its LiDAR research team.
LiDAR has been a crucial technology for Mobileye. It is one of their primary sensors, giving them detailed, accurate 3D maps of the world around their cars. Mobileye’s pioneering work on LiDAR and integrating it within these AI self-driving systems is what made them a world leader in the field. So, why are they abandoning it?
Well, developing LiDAR is far from cheap! In fact, operating their LiDAR R&D team costs Mobileye $60 million a year. Such expenses weren’t a problem when the company was making money. But that isn’t the case any more. Mobileye is expected to post losses for a third year in a row by the end of the year. Why? Well, according to Mobileye, the demand for their products is massively volatile. This volatility is likely caused by the wider automotive industry, which has been drastically changing and struggling recently, and the AI industry potentially reaching a bubble burst point, making investors and customers cautious.
As such, they needed to cut costs, and as such, the LiDAR team had to go.=
Which is a massive problem for Mobileye.
You see, what gives Mobileye’s systems the edge over the likes of Tesla is its “true redundancy”, which requires LiDAR. Its systems actually run two separate self-driving AIs running off different sensors. One uses the LiDAR, Radar and ultrasonic sensors, and one uses the camera array. It would then check the outputs of these AI systems against each other, enabling it to find and ignore erroneous and incorrect outputs. These checks and balances enabled Mobileye to reach driving accuracy and safety levels that are practically unparalleled in the industry. This setup is impossible without LiDAR, which gives almost all the 3D data for one of these AIs.
In theory, you can reach a similar level of self-driving AI accuracy as this “true redundancy” system by just developing a more accurate AI. In fact, this is what Tesla is doing. It chose to save money, using only camera sensors for its self-driving systems. It would make up for this shortcoming by creating an insanely accurate computer vision AI using the millions of miles of data created by Tesla drivers. But in practice, this approach is deeply flawed. As I covered in a previous article, AI is starting to hit diminishing returns, and as such, the costs associated with increasing the accuracy of AI are increasing exponentially. As such, it isn’t even certain that Tesla can afford to create the AI it needs to unlock safe, fully autonomous driving with its current systems.
It seems Mobileye is cutting its nose off to spite its face. Yes, it is saving money now, but without LiDAR, it risks incurring these insanely vast AI training costs down the line.
But Mobileye has a solution.
Firstly, it said it would focus on its EyeQ6-based computer vision perception system, which is almost identical to Tesla’s system. Not promising.
However, it also said it would use cheaper third-party LiDAR sensors. These might not integrate into their systems as well as their in-house sensors, but they would mean they can still use their “true redundancy” setup. They also said they would focus on using imaging Radar, which is incredibly promising. This technology can produce a 3D map using radio waves rather than light like LiDAR does. This makes it cheaper, dramatically expands the conditions in which it can be used in (compared to LiDAR), and can be used over longer distances than LiDAR. Despite not being as accurate as LiDAR, many tests indicate that imaging Radar can identify pedestrians, cyclists and other vehicles better than LiDAR.
So, as long as Mobileye doesn’t fall down the same dead-end rabbit hole as Tesla, its future might still be bright. At least, that is my prediction. However, using cheaper off-the-shelf LiDAR and imaging radar could not only make their systems far more robust and accurate but could also make them significantly cheaper to develop, build and maintain. So, in short, Mobileye still has a shot of staying at the forefront of the self-driving industry.
Thanks for reading! Content like this doesn’t happen without your support. So, if you want to see more like this, don’t forget to Subscribe and help get the word out by hitting the share button below.
Sources: The Register, MRL, Infosys, Will Lockett, Will Lockett