AUSTIN, TX – Technology supplier Aptiv, which last year was spun off from global parts supplier Delphi, may mete out more of its proprietary autonomous-vehicle data, although it has no timetable for the sharing and would withhold the crown jewels.
“The backbone of AV technology, and I mean industrywide, came from the academic community,” says Karl Iagnemma, president-Aptiv Autonomous Mobility and founder of nuTonomy, the AV unit spun out of the Massachusetts Institute of Technology and snapped up by Aptiv for $400 million in 2017.
“And us ex-academics have an open-source ethos,” says Iagnemma, who headed MIT’s Robotic Mobility Group before the acquisition. “We make progress as a community by sharing.”
Iagnemma’s remarks follow a keynote speech at South by Southwest here, and they are newsworthy because in the field of AV development data represents the bricks companies will lay for a path to successful deployment. And Aptiv has massive amounts of AV data. It operates fleets of AVs in pilot programs around the world, generating over the past few years about 2 terabytes per hour over 20-hour test sessions nearly every day of the week.
It would be nearly impossible to place a dollar value on Aptiv’s data hold. But the company is so far along in its AV development, rivals likely would jump at the chance to pinch even a single kilobyte. And if it were free, well, that would be a pot of gold at the end of the rainbow.
Last year, Aptiv tiptoed into data sharing with the public release of NuScenes, an AV data set reportedly containing about 1,000 videos, 1.4 million photographs, 400,000 Lidar scans and another 400,000 image highlights, or bounding boxes, with attributes such as visibility, activity and pose. The scenes were from Aptiv fleets in Singapore and Boston, where dense urban traffic makes for challenging driving situations. The data was annotated by Aptiv's annotation partner, Scale.
“It was pretty valuable,” Iagnemma says. “It takes a lot of time and effort to curate it and mark it up. So, if you’re a developer and you don’t have access to (AVs), which are very expensive, that means you don’t have access to the data and you can’t develop the technology. Practically, you have to have access to the data.”
NuScenes was the largest open-source data set of its kind and gives hundreds of AV-development players, from academic institutions and basement-dwelling developers to Aptiv rivals such as ride-hailers Uber and Lyft, access to information to improve their products. More than 1,000 users registered to access the data, including 200 academic institutions.
“Our belief is, if there are trade secrets, and we’re a company making money, you have to protect the stuff that’s really valuable,” he says. “But you also want to share stuff that enables the entire industry to (improve) safety.”
Safety lapses, such as the highly publicized Uber crash that killed a pedestrian last year in Arizona, reflect poorly on the entire industry at a time when AV developers are trying to win trust in their technology.
“We want to raise the bar on safety, so we’re big believers in sharing and open-source, within some limits,” Iagnemma says.
Another nuScenes release is planned for March.
At the same time, Iagnemma says, Aptiv likely will not reciprocate and seek help from a software development community hungry to get on the company’s radar. Think of it as an it’s-not-you-it’s-me relationship, he suggests, despite the potential power of crowd-sourced solutions from the technical community.
“It’s hard for me to envision a way we could have people contribute to the system in a rigorous and vetted way,” Iagnemma says. “It is a safety-critical system. For the research side of me, that was par for the course.
“You would welcome contributions from individuals everywhere. In our case, the amount of effort you would have to put into vetting the contribution would be so high that it would not be a benefit,” he adds. “I’d love to be proven wrong on that, too, because I think it would be fantastic.”
Iagnemma also tells the crowd Aptiv is researching unique ways to improve perceived safety, including messages to AV passengers in the form of thought bubbles explaining why the vehicle performed an unplanned maneuver, such as evading another car that may have crossed the center line or a pedestrian who stepped into traffic. Or perhaps the car saw something incorrectly and needs to explain itself.
“The car sometimes might be a little bit wrong. It may have thought it was a pedestrian, but it was actually something else,” he explains. “But even if it was the wrong judgment, you’ll be willing to tolerate (it)” with the message.
“It’s performance through testing and it is one of the really valuable (reasons) for doing large-scale testing with members of the public. Getting engineers into the cars, getting our Ph.D.s into the cars, no matter how hard they try it is impossible to put themselves in the point of view of a passenger.”
Among the global, real-world AV testing conducted by Aptiv is a ride-hailing program with Lyft in Las Vegas. Since it launched during CES last year, the 30-vehicle robotaxi service has provided 40,000 rides with a driver monitor.
After each ride, passengers are asked to rate the experience on a scale of 1-to-5. Iagnemma says the average response is 4.95, and nine out 10 riders say they would take another trip in the AV. He says return ridership is strong.
“That tells me everything is being satisfied. They feel comfortable. They feel safe (and) they are getting where they need to go in a rapid manner,” he says.
Aptiv expects commercial deployment of AVs in geofenced areas, such as college campuses, to follow today’s emerging ride-hailing applications in the 2021-2022 timeframe. By 2030, the company expects personal AV ownership to begin if cost curves are maintained. In 2050, significant AV penetration should be seen, the company forecasts.