How Intel and Mobileye are taking a simpler path to autonomy
We take a trip to Israel to ride in Intel and Mobileye's radically simple, surprisingly aggressive autonomous car on the tumultuous streets of Jerusalem.
Video: An autonomous trip through Jerusalem in Intel's ride
While it's likely thatwill be the first company to make driverless cars a commercial reality, thanks to its Phoenix-based autonomous transportation service that launches soon, when it comes time for you and I to actually buy an autonomous car, there's a very strong chance it'll run technology created by Mobileye.
The Israel-based company, acquired by Intel last year for a whopping $15.3 billion dollars, is taking a different, more-direct path to autonomy. While the competition throws millions of dollars at prototype cars festooned with more sensors than landed Apollo 11 on the moon, Mobileye has been relatively quiet as it slowly and iteratively develops its own technology with one presiding directive: KISS. (That's "keep it simple, stupid" for the non-engineers out there.)
Mobileye has created its fully autonomous prototype car using only cameras, a low-cost and (relatively) simple solution built atop the company's decades of experience building adaptive safety systems. While the technology isn't ready for prime time, it is good enough for Mobileye's VP of technology, Shai Shalev-Shwartz, to take me for a demonstration ride around the tumultuous streets of Jerusalem. Hands-off, of course.
Who is Mobileye?
While you may not have heard of Mobileye, I can almost guarantee that you've heard of the company's products. There's a very good chance you've even used them. Nissan's ProPilot Assist? That's Mobileye's tech. Audi's lauded semiautomated system coming ( ) in the new A8? Also from Mobileye, as was most of Audi's previous camera-based driver assistance systems.
Mobileye has provided some sort of technology to nearly all the world's auto manufacturers, but its most infamous white-labeled product was Tesla's Autopilot 1.0. The capabilities of this product, Tesla's initially laissez-faire attitude towards driver attentiveness and the 2016 at the wheel of a Model S vaulted Mobileye into the public eye. That made the all the more dramatic.
While that wasn't a high point for the Israeli startup, the $15.3 billion acquisition by Intel last year, plus the millions of cars already on the road using its technology, indicates that Mobileye is now on the right path. The next goal? Full autonomy... though getting there will be something of a process.
My lap of Jerusalem
To say that Jerusalem is an interesting place is the understatement of the millennium, but what you don't often hear is that it's also a very interesting place to drive. To the Israeli driver, the horn is the most significant signaling device, used far more often than the humble blinker and with little more effect. On the highway, the only way to successfully merge is to prove, though signs of outright vehicular aggression, that you are willing to sacrifice the fender of your car to get that spot.
Aggression is something that human drivers can typically summon to a greater or lesser degree when situation or stress-levels demand, but for an autonomous car, it's a trait that must be acquired.
During my time in Mobileye's car, a retrofitted Ford Fusion, I learned that it is aggressive to say the least. Yes, it indicates properly before initiating any lane change (with the blinker, not the horn), but it also uses no shortage of body language to signal its intentions in a way more clearly understood by the local drivers.
When it was time to merge on the highway, Mobileye's Fusion accelerated without hesitation, charging into gaps and, when no gap was available, creating one by nudging its way over into the lane until one of the other, human drivers decided that discretion was the better part of valor. Gap created, the car promptly filtered in.
That may sound a bit odd and frankly unnecessary in the grand scheme of autonomous behavior, but without this I'd probably still be stuck on the side of an Israeli highway, watching the others stream by and listening to the incessant click of the futile turn signal.
This car's aggression is bounded by a system Mobileye calls Responsibility-Sensitive Safety, or RSS. It's a 20-odd page scientific paper that attempts to define a level of acceptable, safe behavior for an autonomous car. Think of it like Isaac Asimov's Laws of Robotics -- just a whole lot longer and far more complex. Why? Well, an autonomous car could be made incredibly safe in isolation by crawling from point to point, never ever making a left-hand turn and basically behaving in such a way that would make it useless in the real world. In reality, among human drivers, a car must have some level of aggression to survive. RSS defines that boundary between too conservative and too risky.
Interestingly, that level of aggression can be adjusted, and Shalev-Shwartz envisions a time where his company's car will actually tune its own behavior to suit the level of aggression of those around it. Useful, perhaps, for those commuting from the Boston 'burbs up to southern New Hampshire on the regular.
More interesting, though, is how this car sees the world around it.
Sensibly limited sensing
Your average prototype autonomous car is pretty easy to spot, what with the lidar sensors spinning away on some sort of roof-mounted appendage. Even Waymo's car, one of the most subtle of those that can responsibly drive itself, looks like it got a little too cranky during morning recess and now has to wear a dunce cap for the rest of the day.
Mobileye's Ford Fusion, however, is distinctive in its subtlety. At a glance, you'd never notice there was anything special about it. And that's by design. Instead of the unsightly laser scanners that sit and spin away on the roof of most autonomous cars, and indeed without even the radar emitters that have resulted in substantial blocks of flat plastic on the front of many cars, Mobileye's car currently makes do with only cameras.
Mind you, it's a lot of cameras, 12 in total, with 3 dedicated to the task of looking straight ahead, each at a different focal length. Another camera looks back, 2 each on the left and right look diagonally forward and back, while a further 4 are embedded to survey the immediate surroundings of the car. Like I said, that's a lot of imaging sensors but at a cost of $10 or $20 each, you could outfit a whole fleet of cars for what Velodyne will charge you for a single lidar scanner.
They're cheap because they're basic, just 1.3 megapixels each, but the company is upgrading to 8 megapixels soon. That dozen cameras, plus of course GPS and a couple decades' worth of learning from the company's various imaging-based driver assistance systems, is enough to make the car drive itself.
Except that it isn't. At least, it won't be when it comes to delivering the kind of on-road redundancy that Mobileye's engineers demand. And that's why Mobileye is now working on a lidar and radar-based solution. However, unlike most autonomous car developers, which unify their sensor data into a single input for its AI driver, Mobileye is actually keeping things wholly separate. The goal? Full redundancy.
"Once you have a comprehensive solution with only cameras, then you can start pin-pointing where you need additional sensing," Mobileye CEO Amnon Shashua told me. "You are now in a better position to fine-tune exactly where you need a three-sensor modality, and where you need only two... If you fuse from the beginning, for every angle you'll need all three."
So, by adding lidar and radar latter, the other part of the three sensor types Shashua mentions, you can apply them more strategically. Ultimately, you can create two, discrete yet self-sufficient systems, so if either goes offline the other will be able to continue the drive. My ride was fully powered by cameras, but along the way VP Shai Shalev-Shwartz told me that in just a few months, when the lidar and radar system is fully operational, we'll be able to cover the same loop without the cameras.
And, crucially, even when Mobileye does add lidar, the company's cars won't rely on the large, spinning, roof-mounted type of sensor used by most. Instead, it'll use Valeo Scala sensors, box-shaped and designed for integration into the bodywork. Each only scans 140 degrees, so the car will need a number of them to provide full coverage, but at a cost in the "hundreds of dollars" each according to Shalev-Shwartz, that should still result in a remarkably affordable solution vs. the competition's six-figure sensor packages.
Keeping it simple and safe
My ride in Mobileye's car, though fascinating, was ultimately uneventful. That's a good thing, mind you. The biggest missing piece at present, beyond the additional sensors, is the handling of pedestrians. Right now Mobileye's car can detect those walking on foot, and indeed, the in-cabin display highlighted them accurately during our drive, but the system still lacks the necessary smarts to avoid them in a way that is both safe and comfortable. I'm told that will be added soon.
Looming over all this technology, however, is the question of just how a company like Mobileye will get it to the masses. Having partnerships with nearly all the world's major manufacturers means that the tech should have no shortage of potential platforms, and that's a good start. However, the world is still a tangled web of confusion when it comes to legislation surrounding autonomy. That's exactly why Audi's semi-automated A8 will bewhen it hits the US. Without a little regulatory help, Mobileye's new solution will be similarly castrated -- even if it is the best and the cheapest.
What the world needs is some standardized definition of behavior for autonomous cars, a minimum threshold of capability that would define whether a car is "good enough" for public use. For Mobileye, that standard is defined in RSS, and the company now wants it to become the industry standard. Mobileye has opened that document to standards groups in the hopes that others will adopt it. Given how much traction we've seen with other, I fear that well-intended notion may prove overly optimistic.
Still, when the legislation is ready, my money is on the company that kept it simple.
Editors' note: Roadshow accepts multiday vehicle loans from manufacturers in order to provide scored editorial reviews. All scored vehicle reviews are completed on our turf and on our terms. However, for this feature, the manufacturer covered travel costs. This is common in the auto industry, as it's far more economical to ship journalists to cars than to ship cars to journalists.
The judgments and opinions of Roadshow's editorial team are our own and we do not accept paid editorial content.