Assume you run a consultancy focused on AV applied research. In the pandemic era, where trade shows have all gone virtual, how do you strut your stuff?
Assume you run a consultancy focused on applied research in AV (automated vehicle) technologies. When practically every tech (or automotive) conference has been either postponed or gone virtual, how do you strut your stuff?
“You’ve got to bring the show on the road,” said Phil Magney, VSI Labs founder and principal advisor. Magney last week stopped by in Madison, Wis. for a chat with EE Times. He and Matt Linder, VSI Lab’s lead AV solutions engineer, were on their way back home from Detroit, where they spent three days conducting the intensive testing of autonomous emergency braking (AEB) applications at the American Center for Mobility (ACM).
ACM, which was closed due to the ongoing pandemic, re-opened a few weeks ago. “We were one of the first wave of companies to use the test facility,” Magney explained.
In 2019, VSI Labs researchers completed a 2,000-mile journey from their Minneapolis, Minnesota home to Santa Clara, California, in a road trip initiative called “Automated Drive West” Series. The goal was to examine different ADAS technologies in real-world conditions.
Since this year’s lockdown, VSI Labs is expanding its “Automated Drive Series,” conducting four quarterly driving trips, coast to coast, and border-to-border. Magney explained that aside from work it does for clients, VSI is taking test vehicles on the road to expose ADAS technology to a variety of environmental conditions and collect data.
In addition to testing at ACM, Magney and Linder also used the opportunity to meet clients in Detroit. Given that most offices of car OEMs and Tier Ones remain closed to visitors, this wasn’t exactly business as usual. In one case, Magney said that a meeting took place on the driveway at a client’s home.
The tests VSI Labs conducted at ACM included a technology called AEB with pedestrian (AEB-P). The team did this at the request of VSI’s client Flir, a thermal imaging camera company.
AEB-P, an ADAS feature, is known to be problematic for most ADAS vehicles to handle. Last year AAA conducted tests on a variety of ADAS vehicles, specifically focused on pedestrian detection. The test results published by AAA were devastating.
They verified that AEB-P is a high bar for most vehicles to clear, largely because it is a “collision mitigation system.” While most ADAS features are designed as a collision warning system for drivers, a collision mitigation system is set up to alert the driver. If no action is taken, the system will hit the brakes to avoid or lessen the severity of the collision.
The AAA’s tests found that if an adult was encountered crossing the road in daylight by a test vehicle going 20 mph, the car avoided hitting the pedestrian only 40 percent of the time. Worse, if the test vehicle traveling at 20 mph saw a child darting into traffic from between two cars, the kid got nailed 89 percent of the time. At 30 mph, none of the test vehicles avoided a collision. What about an adult crossing the road at night? Forget about it. The pedestrian detection systems proved ineffective.
So, what’s the issue? Does a limited field of view of cameras make it harder for ADAS vehicles to detect pedestrians? Or does this mean that even the combination of RGB cameras and radar is still not good enough for AEB-P, if used under various environmental conditions?
When ADAS Goes People-Blind
Flir’s answer is to moot the question by adding thermal imaging technology. It claims that thermal cameras can provide “complementary data to RGB cameras and radar.” Earlier this year, Chris Posch, Flir’s engineering director for automotive, told EE Timesthat thermal cameras “see” heat, allowing Flir cameras to detect pedestrians in challenging conditions including night, through sun and headlight glare and fog.”
Magney explained to us, “Thermal cameras have been used for automotive night vision for some years.” But the thermal imaging has yet to be applied to pedestrian detection. “At VSI, we have been working with Flir for over a year to implement AEB-P (Autonomous Emergency Braking Pedestrian).” VSI did testing “on the premise that thermal cameras that are trained at detecting pedestrians will outperform existing visible cameras + radar systems,” he noted.
So, did they ace the tests?
Flir plans to announce its test results at ACM at a later date.
However, Magney explained that VSI tested the Flir-based AEB-P system against a variety of competitive solutions in the market today. Without naming names, Magney said, “They are thought to have the most sophistical ADAS features.” He also added that all tests were done under strict NCAP guidelines and procedures.
A key element in the AEB-P testing done by the VSI team at ACM is the software. Magney explained that they used the solution leveraging the neural networks that were trained for thermal. He noted, “Furthermore, we coupled the thermal classification with object data from the Aptiv ESR LR/MR radar.” Together with a Flir visible camera equipped with a visible image classifier, this AEB-P solution relied on combined data from the thermal camera, visible camera, and radar.
Ruggedized computer in VSI’s test vehicle
Having Magney drop in my hometown was a treat because I got to see firsthand VSI’s testing vehicle — based on a 2018 Ford Fusion. It contains a full stack of sensors that include vision cameras, thermal imaging, LiDAR, radar and precision localization. More important, before embarking on this trip from Minneapolis to Michigan’s Upper Peninsula, Magney’s team installed a ruggedized computer called Crystal Rugged.
Magney explained that the computer is designed for “the toughest and harshest conditions known to vehicle deployments.” Under the hood, you can even see a water-cooled system. He acknowledged, “This is not designed for series production, far from it, but it can withstand very harsh environmental conditions.”
The Crystal Rugged computer, designed for ADAS/AV development activities, comes with dual Nvidia GTX2080i accelerators that can handle “the most complex inference models,” noted Magney. “This is the most powerful computer VSI has every worked with.”
He noted that thermal management and power conditioning keep this system “on a even keel” in the harshest conditions. In fact, VSI was testing AEB applications for three days at ACM in 90-degree heat with track temperatures around 112 degrees. “We were running nonstop. Over the course of each day we executed about 100 AEB maneuvers. And these are hard stops! You can imagine how hard this would be on normal computers,” he noted.
One of the advantages of getting on the road is the opportunity for “data collection.”
Describing the team’s latest trip, Magney said, “We’re focusing more on data collection instead of proving that the technologies work. This is especially important to developers of ADAS and AV gear.” Magney said that these developers “need greater exposure to scenes and scenarios. They are also looking for more edge cases.”
Magney boasted, “With our data collection, we amass a massive scenario database that can be used for testing or training algorithms or optimization of sensing technologies. We accumulate about one Terabyte for every 1,000 miles traveled on average.” Such data can be shared with VSI’s clients, he noted.