Regulation Spawns Software-Defined Radars

Will the new federal regulation mandating automatic emergency braking both in daylight and nighttime on all passenger cars and light trucks "stifle innovation"? Let's break it down.

As the new administration rolls in, deregulation is all the rage.

Traditionally, private companies oppose government oversight, typically citing the decades-old myth that “regulation stifles innovation.”

However, not all regulations are created equal.

High-minded goals stipulated by regulations have, at times, prodded industries toward breakthroughs both surprising and profitable. Regulation can be “the mother of invention.”

Take Federal Motor Vehicle Safety Standard (FMSS) No.127.

Finalized last year by the U.S. Department of Transportation’s National Highway Traffic Safety Administration (NHTSA), this rule forces automakers to feature automatic emergency braking (AEB) including pedestrian AEB both in daylight and nighttime on all passenger cars and light trucks by September 2029.

The regulation addresses vehicles’ nighttime visibility challenge. While AEB has been proven effective, “the autobrake performance slips after sundown,” according to the Insurance Institute for Highway Safety (IIHS). According to NHTSA’s 2020 data, the nighttime fatality rate is triple the daytime rate, and 76 percent of pedestrian fatalities occur at night.

Meanwhile, however, a lawsuit recently filed in the U.S. Court of Appeals for the D.C. Circuit by a group called “Alliance for Automotive Innovation” against the Biden administration’s Department of Transportation seeks repeal of NHTSA’s new AEB rule. It challenges the rule's technological feasibility, asking the agency for a “rule that maximizes driver and pedestrian safety.” The group wants the Trump administration to repeal or revise the AEB rule.

For sure, FMVSS No. 127 sets the safety bar high for automakers. The new mandate is neither simple nor straightforward for carmakers to follow. They need to choose technologies intelligently while solving the complexities of system integration. And they must be ready to swallow additional costs.

Phil Magney, founder and president of VSI Labs, noted that the new mandate will be “a heavy lift” for most carmakers.

“Early [AEB] systems relied on radar and only stopped for vehicles (not for pedestrians). Later camera systems enabled pedestrian AEB but most struggled at night,” explained Magney.

Clearly, either sensor type alone is not enough.

Technology choices
Not everybody believes that this upgrade is out of reach.

For automakers to innovate their way out of the FMVSS No. 127 conundrum, they must first consider adding a new type of sensor, like thermal (i.e. Teradyne Flir, Owl Autonomous Imaging) best for night-time vision.

The second option is to look for a new generation of higher-performance imaging radar (i.e. Arbe, Uhnder, NXP Semiconductors, Texas Instruments). Everyone agrees that one advantage of radar is clear: It works in all weather conditions. On the other hand, it’s difficult for these radars to classify objects as precisely as a camera. There’s also a persistent problem with false positives.

A third option is to deploy smarter camera-radar fusion solutions enabled by high-performing SoCs from companies like Ambarella and Mobileye.

But that’s not all.

“Software-defined” radars are on the horizon for automakers looking for “lower cost” and “scalable” solutions to meet the mandate of FMVSS No. 127. These hardware-agnostic algorithms, designed to apply to radars, are promising.

The challenges created by the new AEB/P-AEB rule have opened opportunities for startups focused on software development that can be layered on top of existing hardware.

Startups such as Neural Propulsion Systems (NPS), a Pleasanton, Calif.-based startup, and Perciv AI, spun off from Delft University of Technology’s Intelligent Vehicles Group, are charting a course into the brave new world of “software-first” radars.

The rise of software-defined radars dovetails with “the transformation we are already seeing in the automotive industry — a shift to software-defined vehicles,” said Pierrick Boulay, senior analyst at Yole Group.

Given that the new regulation applies to an entire range from low- to high-end models, software-defined radars could answer some cost concerns shared by OEMs. While high-end vehicles might be able to afford higher-performance sensor fusion processors, software-defined radars can target “entry to mid-tier cars” through a lower-cost implementation of FMVSS No. 127, observed Hassan Saleh, senior RF technology and market analyst, at Yole Group.

NPS
NPS, established in 2018, is promising a hardware agnostic software-defined radar. CEO Behrooz Rezvani noted that today, all automakers are focused on radars.

If cameras can’t discern objects in bad weather, they won’t get any better by adding horsepower to camera processing, he explained. “You need radar. But traditional radar systems have been highly unreliable.”

Rezvani believes his company is well-positioned, with software-defined radars that, he says, will be able to offer clearer resolution, enable faster detection time, and reduce false positives.

The heart of NPS’s technology is a mathematical framework called “Atomic Norm Software.” Referring to work that “came out of Caltech and MIT,” Rezvani explained, “it has taken us seven years to get here.”

Combining this unique mathematical model with physics, Revani claimed, “we can detect a pedestrian next to a car a tenth of a mile.” By comparison, radar that detects a pedestrian too late — or not at all — won’t prompt the car to slow down in time. Pop goes the pedestrian.

Black box?
Of course, how NPS’ atomic norm formulation works and the exact math behind it, are still secret.

Yu Yang, principal analyst for automotive semiconductors at Yole, said, “Since we don’t know how they achieve it, to us, this is still in the black box.” Egil Juliussen, a veteran automotive industry analyst, agreed. “NPS is too close-mouthed… and apparently, they don’t need to talk to customers,” said Juliussen. This might be because “NPS appears to be already closely working with General Motors.”

It turns out that Larry Burns, GM’s former corporate vice president of Research and Development, vetted NPS’s technology and recommended it to Rick Wagoner, a former GM CEO who is now an investor. “That sort of carries some weight,” said Juliussen.

Last year, NPS’ $17.5 million in series B funding was led by Cota Capital with key investment from GM ventures and RTX ventures.

Perciv AI
Perciv AI approaches reliable radar perception from a slightly different angle.

While NPS strives to improve radar data, Perciv AI’s mission is to take sensor output and make its data “digestible” by vehicles, said Andras Palffy, Perciv AI co-founder. “Our job is to ‘use’ the radar data better.”

Current-generation radars offered by tier-one suppliers suffer from ghost objects and misclassifications. Perciv AI claims that its AI-driven radar perception software running on the same radar hardware can generate a clean, actionable output.

But when the initial radar sensor output is bad, how do you avoid the “garbage-in and garbage-out” syndrome? What processes are necessary to make not so good radar data digestible?

“We take multiple steps,” said Palffy.

First, he said, we have very advanced point-cloud segmentation. Given dirty data, Perciv AI’s software detects and estimates which points are probably noise (or ghost targets) bouncing around in space, he explained. Perciv AI’s networks can filter those out.

The software also automatically estimates which parts of a scene are static and which are moving. It measures the Doppler rate, filtering between objects moving at different velocities.

Finally, Perciv AI extracts features at the point-cloud level, noted Palffy. “As we start thinking at the scene level, we argue what it is that the sensor is seeing. We take into consideration types of reflections, power, and kind of velocity measured…”

The company claims that its software outperforms traditional radar data perception, with help from AI methods and data. While the software has gone into series production in drones, automakers have yet to sign up.

Perciv AI, based in Delft, is still a very young startup. It secured 2.5 million euros in seed funding last month, led by DayOne Capital and Keen Venture Partners, joined by Vinci Venture Capital.

The power of software in the automotive ecosystem
Software companies such as NPS and Perciv.AI will bring new wrinkles to the automotive ecosystem.

Unlike the traditional hardware-centric auto industry, where the supply chain tends toward a vertical structure, hardware-agnostic software vendors can work with different types of automotive players across the ecosystem.

“There are multiple ways to skin the cat,” said the NPS CEO. Licensing agreements can be executed with either tier ones or OEMs directly, he added.

Perciv AI CEO Palffy noted that Perciv AI will be talking to not only system companies but also semiconductor vendors who make radar chips, for example. “Once we work with one, thus helping improve the radar performance, the other company wants it too.”

Given that FMVSS No. 127 will be enforced starting in 2029, vehicle designs that meet the regulation must be finished two or three years in advance, which means 2026, explained Yole’s Boulay.

NPS, apparent with a head start among software-defined radar developers, is scheduled this year to make their “A sample” available “to any and all OEMs or tier ones in the first quarter,” explained the CEO.

Meanwhile, Perciv AI currently has six proof-of-concept projects percolating in the automotive field. Tier ones and semiconductor companies are also evaluating Perciv AI’s software, Palffy noted.