Qualcomm came to the Consumer Electronics Show with one clear message: The cellular phone chip giant is ready to rumble in the high-stakes battle of highly automated vehicles.
LAS VEGAS — Qualcomm came to the Consumer Electronics Show with one clear message: The cellular phone chip giant is ready to rumble in the high-stakes battle of highly automated vehicles.
At CES, Qualcomm was prepared to sneak — but only for a selected audience — a peek at what its “stealth AV program” entails. Of course, once a company starts talking about a “stealth program,” it’s not exactly stealthy anymore, is it?
Obviously, what was said in Vegas will not stay in Vegas.
There are three things to keep in mind about Qualcomm’s AV strategy:
Qualcomm sees connectivity as “the obvious cornerstone of any autonomous strategy,” said Ian Riches, director of the automotive electronics service at Strategy Analytics.
Qualcomm is seeking differentiation from competitors through “its 5G technology and the success of what it is achieving with C-V2X,” said Riches. Qualcomm and Ford, for example, announced during CES their collaboration on a “C-V2X Global Initiative.” Their evident objectives are “to improve vehicle safety, traffic efficiency, and support for autonomous driving.”
Beyond connectivity, though, it remains unclear what Qualcomm has up its sleeve for the AV platform. “Qualcomm obviously has expertise in designing powerful SoCs, such as Snapdragon,” Riches said. “Although, in and of itself, such a processor is probably not what is required …”
In Riches’ opinion, the fully-fledged AV will need, alongside Snapdragon, “some form of neural-net accelerator to form a viable platform.” He referred to a reported Tesla leak that “AutoPilot V3” will be based on a Samsung Exynos SoC (broadly similar in power/architecture to Snapdragon) alongside a Tesla-designed accelerator.”
One Qualcomm edge is time. Now that carmakers are shifting attention from fully automated driving to driver-assistance systems, Qualcomm, in theory, is more likely to find programs into which they can push their technologies.
“I was impressed … as [the company] hasn’t even officially entered the self-driving technology market,” said Patrick Moorhead, president and principal analyst at Moor Insights and Strategy, who was one of the lucky ones to get a ride in Qualcomm’s test vehicle in Las Vegas. “It reminded me of other experiences I have had inside L2- and L3-enabled cars.”
What’s inside Qualcomm’s AV?
First, let’s talk about Qualcomm’s AV test vehicle. As it turns out, the cellphone chip giant already has a DMV license to drive test AV cars in California. It also has permission in Nevada, which allowed Qualcomm during CES to joyride a host of VIPs in its robo-car.
So what are the building blocks of Qualcomm’s AV?
Inside Qualcomm’s autonomous vehicle are seven cameras providing a 360° surround-view, radars both long-range for front and rear and short-range in all four corners, a global navigation satellite services (GNSS) antenna for positioning, an in-cabin camera array for Vision Enhanced Precise Positioning (VEPP — more on this later) and forward-view perception, and Qualcomm’s Drive Automotive Platform based on its Snapdragon Automotive Processor.
This test vehicle does not come with a LiDAR. Anshuman Saxena, director of product management, ADAS, and autonomous driving in Qualcomm’s automotive business unit, made clear: “We are not saying that LiDARs aren’t necessary. There are roles that LiDARs can play. However, our focus is on providing OEMs with [autonomy] technology solutions at the right cost point.” In Qualcomm’s view, car OEMs can build advanced ADAS cars without LiDARs.
But what levels of autonomy do Qualcomm’s AV testing vehicle provide? Saxena declined to comment. He told us, “We do not label it as Level 2, Level 2+, or Level 3.” He described what the company offers as “self-navigation highway autonomous driving.”
Qualcomm’s AV is capable of “maneuver and trajectory planning” on highways, he explained, in addition to the usual lane-keeping and automatic emergency braking. Object-level fusion using cameras and radars will result in obtaining “classification, range, and velocity,” which the AV needs for behavior prediction and behavior planning, according to Qualcomm.
For example, behavior prediction of other vehicles is enabled by “motion models and blinker/brake detection.” The AV is capable of behavior planning including “lane keeping, yielding to cut in, and lane changing,” the company said.
Qualcomm’s AV is not designed for fully autonomous driving. It requires a driver to take over when needed.
Qualcomm’s AV today is based on the “Qualcomm Drive Automotive Platform,” using four modules, each based on the company’s second-generation Snapdragon Automotive Processor. This is a platform designed to be scalable. To fit the autonomy level that car OEMs need, modules can be added, as shown below.
Qualcomm's Drive Automotive Platform (Photo: EE Times)
Qualcomm, currently working on the Gen 3 Snapdragon Automotive Processor, can put together the same AV by using only three modules in the platform, said Saxena. The company, however, isn’t yet revealing details of its Gen 2 or Gen 3 processors.
Qualcomm doesn’t make radars, but by laying deep learning atop commercially available radars, the company believes that it expands the radars’ performance. For example, by using Qualcomm’s internally developed “Radar Deep Neural Network,” the company claims that the AV can extract rich features from raw radar signals. A 2D bounding box can estimate and classify vehicles on the road, perceiving the difference between a truck and a car, for instance. Qualcomm, however, hopes to eventually get higher resolution and 3D bounding boxes by using enhanced radars.
Qualcomm’s AV goals
Qualcomm has been working on its AV platform for about three years in stealth mode, said Saxena. Qualcomm’s goal is to “leverage a lot of assets we already have” in the company’s AV initiative and provide the automotive industry with cost-efficient and power-efficient solutions.
The backbone of the company’s AV technology has four elements: localization, cameras, radars, and cellular V2X. Qualcomm is focused on enabling a lot of sensing and planning features so that, “ultimately, the car sees everything around us and it can predict what is going to happen in the next two to three seconds,” said Saxena. Qualcomm’s mission, he said, is the pursuit of efficiency. By adding autonomy, “we hope to free up drivers’ time as much as possible.”
That, however, doesn’t mean that Qualcomm is rushing to develop solutions for Level 4 or Level 5 vehicles. Qualcomm is speaking of “efficiency” in terms of an advanced ADAS vehicle that can park itself in a garage, for example, after a driver gets out.
What is VEPP anyway?
Perhaps the most intriguing technology that Qualcomm has been developing in its stealth AV program is what it calls “Vision Enhanced Precise Positioning (VEPP).” VEPP is a unit that “fuses multiple automotive sensors — GNSS, a camera, an inertial measurement unit (IMU), and wheel sensors — to offer more accurate global positioning of vehicles,” according to Qualcomm. Saxena believes that VEPP will prove to be “a very cost-effective solution for car OEMs” because it uses sensors already present in today’s vehicles.
VEPP does not depend on the availability of any kind of map.
Asked about Qualcomm’s localization technology, Phil Magney, founder and principal advisor at VSI Labs, described VEPP as “a cross between a telematics control unit and a high-end positioning device.” He called it “a self-contained ECU that couples connectivity with correction including visual odometer, inertial navigation systems, and, now, landmark-based localization.”
VEPP “would fit well into series-production vehicles with active ADAS and automation features because you can increase the safety and performance and enable some features that you could not without some precision localization,” said Magney. “Some examples are heightened lane guidance, intersection traversal, and navigation in and around dense buildings.”
Specifically, Qualcomm claims that VEPP testing achieved “lane-level accuracy — accurate to < 1 meter.” In contrast, today’s GNSS-based solutions can achieve similar results only in open-sky conditions and by incorporating IMU and wheel sensors, according to Qualcomm.
Another VEPP advantage is that it works in challenging urban environments. Qualcomm also said that VEPP’s accumulated error over an entire drive route in a parking garage is “on the order of 0.1% of the distance traveled.” The company added, “This represents significant improvement over dead-reckoning solutions available today.”
Magney told us, “This is clever. And stealth.”
However, he added that Qualcomm has been working on this for a couple of years. “They showed it last year at CES (though VEPP is a new acronym to me).” Describing Qualcomm’s AV program as the third-generation Snapdragon platform that does C-V2X (connectivity), infotainment, navigation, etc., he noted, “Now it is optimized for landmark-based localization. This is very similar to Mobileye REM and Nvidia Drive Localization. The purpose is to build and maintain a localization layer that resides on top of a precision map.”
Magney remains cautious in his assessment of VEPP. “We don’t now know how well this works as we have not tested it.” But Magney is intrigued. “Normally, offering precision localization requires some costly hardware such as differential GPS or real-time kinematic (RTK) correction,” he observed.
Strategy Analytics’ Riches remains skeptical of VEPP. He told EE Times, “There’s still a lot of uncertainty over mapping and location requirements. I don’t see VEPP as a game-changer for Qualcomm.” He said that it’s more a “potentially useful piece of the jigsaw at the moment.” He added, “We are still seeing a lot of interest in enhanced GNSS techniques, such as the Trimble RTX solution that is used by GM for SuperCruise.”
Qualcomm’s AV future
Overall, the analyst community is favorable toward Qualcomm’s AV program. Egil Juliussen, director of research for ADAS at IHS Markit, believes that Qualcomm’s methodical approach toward ADAS and AV will prove to be just as successful as the company has been with its modem chips now designed into most of the telematic units in vehicles. “It took them a few years, but Qualcomm is using the same strategy for cockpit, ADAS, and autonomous vehicles.” Qualcomm’s scalable automotive SoC solutions — learned from its smartphone business — and lower-power chip solutions likely to be equivalent to those of Mobileye will be a good asset for Qualcomm, he explained.
Both Juliussen and Mike Demler, senior analyst at the Linley Group, agreed that one Qualcomm advantage is time.
“Coming out of CES, I see many others in the industry are starting to step back a bit from the mad rush to ‘full autonomy,’” Demler told us. “There is more of a recognition that high-level autonomy is a long-term R&D project, at least for mass deployment, regardless of the early robo-taxi projects. So Qualcomm may be behind on the flashy Level 4/5 projects, but nobody is making any money off of those yet anyway. The volume will be in Level 2 (or the new Level 2+) for quite some time.”
Juliussen, however, noted one caution. “I wonder about the ecosystem … how will Qualcomm be able to build an ecosystem around the company’s solutions among tier ones and vendors who manufacture automotive boards?” Given that Mobileye/Intel and Nvidia are further along in ADAS and AV partnerships, Juliussen said, “It’s always hard to kick out incumbents, especially in the automotive industry.”
In other words, it’s not easy to transplant solutions already designed into vehicles, “unless you find they are doing something wrong in their solutions.”
One more thing. Asked about Qualcomm’s future in the automotive industry, Strategy Analytics’ Riches said, “Qualcomm’s biggest weakness remains its (relative) lack of functional safety expertise.”
In a separate EE Times interview, Qualcomm senior vice president Nakul Duggal stressed that the company learned hard lessons about what it takes to be an automotive chip supplier. Duggal went so far as to claim that automotive changed the company’s DNA.
— Junko Yoshida, Global Co-Editor-In-Chief, AspenCore Media, Chief International Correspondent, EE Times.