Sensors Step Out While Startups Stutter
Highlights from the 2022 AutoSens conference show technical progress and confirm some industry suspicions.
The Michigan Science Center in downtown Detroit was an appropriate venue for the recent 2022 AutoSens conference and exhibition, given the Science Center’s key role in promoting interest in engineering, physics and the natural sciences. At this year’s AutoSens, attendees heard from more than 40 expert speakers about the latest in automated-vehicle perception technologies, OEM development progress and the volcanic sensor market. Some highlights of the event:
Cruising toward second-gen Origin
“The benefits of autonomous vehicles can only be realized with [production] scale,” asserted Shane McGuire, lead principal systems architect at Cruise LLC, GM’s AV development organization. In his keynote presentation at AutoSens, McGuire affirmed that commercial fleets and ride-hail companies are the vanguard of driving SAE Level 4 driving automation toward higher volumes he says will arrive by 2025. Cruise has been landing fare-paying ride-hail business in San Francisco using its AV-configured Chevrolet Bolt platform with driver/observers in the vehicle for late-phase systems validation. Early validation work was done with simulation.
The purpose-designed Cruise Origin roboshuttle vehicle is on track for production in 2023, while development of a more advanced second-generation Origin continues. “We’re completing the R&D,” McGuire said, “and co-developing the vehicle with its sensor suite.” The next Origin’s electronics architecture is divided into “pods,” each with its own sensor array.
“We’re consolidating the sensor suite with modular interfaces and a centralized architecture –sensors, compute, telematics — using Cruise-customized SoC [silicon on a chip],” he told the AutoSens audience. “The aim is to do as little processing at the ‘edge’ as possible; to push processing to a central processor. And we are tailoring each sensor to what it’s best at” to reduce sensor redundancy — and consequent systems cost. MLA (micro-lens array) semiconductor technology, enabling huge data storage, “forms the bulk of our software stack, and is our core processor,” McGuire explained.
“Finding silicon for a specific company’s AV stack is difficult,” he noted, acknowledging that custom silicon development can take up to six years. Near- field lidar, being developed in-house by Cruise, also has required customized silicon.
McGuire expects the first-gen Origin, at launch, to operate at a cost of $4.25 per mile, an uncompetitive rate that he believes will be progressively reduced through systems design — and particularly systems consolidation.
He used the AutoSens venue as an opportunity to solicit the supplier community for collaboration, with particular focus on three technologies: high-resolution radar, nearfield lidar, and camera systems. Radar technical requirements include all-weather capability, 4D imaging, and 360-degree field of view (FoV). Cruise’s nearfield lidar currently in development will have 3D high-resolution output, with grayscale intensity and range being premier performance targets. Camera systems must have high resolution, superior low-light performance, 360-degree FoV, and “long range at minimal systems cost,” McGuire offered. Software solutions are key: “Processing high-res camera data is exceptionally expensive.”
Omnivision CTO lays out challenges
AutoSens 2022 coincided with two major product announcements by Omnivision: the new OAX4600 AI-enabled ASIC for simultaneous driver- and occupant-monitoring systems; and the company’s partnership with Valens Semiconductor on a MIPI A PHY-compliant camera solution for automotive applications. The companies will include Valens’ new VA7000 chipsets inside Omnivision’s automotive reference design system (ARDS) camera modules.
Omnivision CTO Dr. Boyd Fowler, told SAE Media that while camera applications continue to proliferate, “everybody wants better low-light performance [<1 lux], and more pixels within the same format,” with pixel size being driven down to 2.1μm. Fowler unabashedly laid out other challenges in the automotive space.
“LEDs are pulse-width modulated, making flicker from road signs, billboards and headlights huge problems” for optical sensors,” he noted. “The challenge is in achieving wide dynamic range with good flicker mitigation. Omnivision 3-exposure/4-capture Split Pixel HDR (high-definition radar) was developed to mitigate LED flicker issues, he said. Fowler also detailed another automotive customer request, image sensor junction thermal performance up to 135C (275F) — “very difficult to achieve,” he opined, explaining that automotive low-power operation typically calls for 20-to-40 deg. C (68 to 104-deg. F) package thermal resistivity.
The new OAX4600 integrates a unique feature set, compact form factor, and low-power requirements, with ASIL-B functional safety, specifically developed for vehicle cabin monitoring systems, Fowler said. The chip uses Omnivision’s ISP processing and dedicated NPU to ensure robust, latency-free image processing even when faced with RGB infrared. The chip is capable of higher resolution processing of up to 5 megapixels and features integrated cybersecurity.
Regarding cyber, Fowler highlighted concerns including fake and/or counterfeit sensors or ASIC chips installed in vehicles by hackers, and RFI jamming. “Everybody wants cybersecurity but nobody wants to pay for the computing,” he stated.
Woodside’s Burger feast
No automotive tech conference is complete without insights from market analysts. In the case of AutoSens 2022, the analyst was Rudy Burger from Woodside Capital. His compelling presentation included the following nuggets:
- As of May 2022, there were 238 companies operating, or claiming to operate, in the automotive perception space
- “Expectations of the corporate investors that are OEM-affiliated [i.e., GM Ventures] are rising steadily each year. They want to see pre-production units” before investing in new startup tech companies, Burger noted. And the VCs also are getting tougher to raise capital in the sensor space
- “There are about 100 lidar ‘companies’ now. But consolidation is underway and I expect that total to be reduced to about five companies
- “It is now very difficult to get a new lidar company funded unless they’ve got really special technology,” he said; 2016 was the peak year for lidar startups (14), declining to 4 startups in 2018 and only one in 2021
- The top four best-funded automotive perception companies are Chinese
- “I see very little correlation between the amount of capital a company raises and their eventual commercial success”
Bright Way shares its ‘Visdom’
Israel-based Bright Way Vision was founded in 2011 as a spinoff from Elbit Systems, a key defense- tech company. At AutoSens, Bright Way co-founder and executive VP Eyal Levi, and the company’s head of business development Meny Benady, shared their “Visdom” with SAE Media about what they claim is the world’s first automotive night and all-weather camera system (called Visdom).
“We’ve spent a lot of time with the biggest global OEMs, all of whom tell us that low-vision and extreme weather driving scenarios are the greatest challenges in developing their automated-driving systems,” Levi said. “They must have camera systems that deliver outstanding performance day, night, and in all weather conditions, along with sensor redundancy. And it must be low cost.”
The Visdom product is a CMOS-based all-weather vision system powered by Bright Way’s patented GatedVision technology. Visdom uses Class-1 (invisible to the naked eye) near-infrared (NIR) illumination pulses to generate ‘slices’ of the space ahead of the camera lens, while thousands of synchronized micro-exposures of the gated camera build a complete, continuous, illuminated image of the road ahead.
The compact Visdom sensor unit can be mounted behind the windshield or integrated into headlamp assemblies. It is available in three configurations for passenger vehicles, trucks, light trains, buses and robotaxis. Levi noted that Visdom delivers a uniform illuminated range of up to 300 m (984 ft.), horizontal FoV from 16 to 60 degrees, and resolution up to 2M pixels. The high resolution of the images created by GatedVision sets Visdom apart from typical automotive cameras that struggle to collect images in poor visibility conditions, Levi said. Visdom is currently in serial production, he said.