Bosch, Continental, Denso, GM, Nvidia, NXP, and Toyota declared their support for the new consortium.
During Arm TechCon in Santa Clara this week, a group of automotive and technology companies, including Bosch, Continental, Denso, General Motors, Nvidia, NXP Semiconductors and Toyota, declared their support for the newly launched Arm-led Autonomous Vehicle Computing Consortium, Inc. (AVCC).
The group, organized quietly over the past two months, plans to develop a set of APIs that sits above underlying autonomous vehicle (AV) hardware, Armando Pereira, President of AVCC, told EE Times. “Our focus is predominantly on hardware.”
The consortium’s goal is to optimize semiconductor solutions by writing a standard set of common requirements for hardware. “Those in the AV industry are seeing a huge opportunity in it, because they want to develop AVs that are less power hungry, and that can be manufactured at scale,” said Pereira.
Phil Magney, founder and principal at VSI Labs, sees the formation of AVCC as inevitable. It signals that “developers, OEMS and even suppliers are finally agreeing that collaboration may be necessary.”
Today, designing and developing an AV stack is a tough task because a thousand combinations of hardware produce the same results, he explained. Calling an AV “a system of systems,” Magney said AV designers must deal with so many processor types, memory types and supporting logic, “Who’s to say they got it right?”
In other words, a confluence of future uncertainties has pressed key players to see the value in flocking together.
“Too many wildcards” and “too many expensive bets” have already characterized the fledgling field of AV. Asked about AVCC members, Magney said, “All the companies that are members are well along in their own development of AV stacks and have already invested millions but there is no guarantee. Furthermore, many of the largest companies, including those initial members, may not be satisfied with their efforts thus far. Even GM and Toyota who have invested billions are not certain their solutions will prevail.”
Magney observed, “Up until this point the AV stacks out there are bloated compute platforms that draw too much power, generate too much heat, and are too big and bulky. Just look in the trunk of any AV development platform!”
When any industry consortium like AVCC is born, the list of members is the metaphorical shiny object, but it's important to pay attention to who’s not in the group. The absentees include Intel/Mobileye, BMW, Qualcomm, Waymo, Ford-Argo, Uber, Tesla and Volkswagen.
AVCC’s president Pereira made it clear that there are no limits to AVCC membership. Given that the consortium is still in its early days, he said, “We are not concerned” about its smallish numbers. He expressed confidence that more OEMs and Tier Ones will sign up.
Magney told us, “I don’t see Intel jumping on the bandwagon just yet. AVCC does lend itself to Arm because the objectives of AVCC are core to ARM IP.”
The consortium is where Arm can tout its own cores’ ability to offer such advantages as “high performance, low power and proven ability to scale.” Magney added, “Also, many of the companies listed already license Arm processors as they are well-suited for host, real-time functionality. Nvidia and NXP, for example, use Arm cores in their SoCs designed for automated driving.”
Details yet to be defined
Pereira indicated that the group will be working on such areas as “interfaces with sensors” and “data flow” of images inside AVs. Asked about details, he said the group is not ready for that discussion.
On sensor fusion, for example, an unresolved issue is whether AVs should do it early or do it late. With an AV industry still wildly split in its approaches to system architecture, it’s not clear how the AVCC can coax its members into any agreement.
Acknowledging that the group will be working on sensor fusion recommendations, Pereira noted, “None of us is in a position” to describe the outcome.
The AVCC is aiming to produce work product by late 2020. Until then, how effective “system architecture” solutions AVCC might produce remains a mystery.
Magney is remaining optimistic about the AVCC collaboration. He believes the goal of the group is to get input from members who can collectively define the compute requirements. “A lot of experience comes to the table to sort out the technical requirements. I think this gives the members another architectural approach but does not preclude the members from their existing designs.”
Since the group just became public, the AVCC website lists no specifics either on its committees or working groups. Pereira, however, told EE Times that the chair of the AVCC technical committee, for the time being, is Paul Hughes, Arm’s system architect for automotive.
Hardware vs. software
Clearly, the fragmentation of AV hardware architecture is not helping car OEMs build AVs that can scale. But it’s often argued that AVs will live or die by AV software, not necessarily hardware. Why focus on hardware first?
Magney agreed that software matters more. “Let’s solve the software requirements then define the target computing platform that best meets those recruitments.”
Is AVCC “putting the cart before the horse?” Answering his own rhetorical question, Magney noted, “Not really.”
He explained that the AVCC does not call for specific IP but attempts to create a reference design for the full computing pipeline within an AV stack. “Developers who use AVCC can specify their own processing that meets the requirement of their SW stack,” Magney stressed.
Pereira said that software is out of the AVCC’s scope. Software is being handled by another industry group, he added.