The driverless Toyota Sienna minivan came to a halt in a parking lot when it detected a garbage truck blocking its path.
A human driver might have noticed that the garbage truck hadn’t started collecting trash from the nearby dumpster and opted to go around the vehicle.
But the autonomous vehicle played it safe. It waited for the truck to finish its job and move.
“The vehicle might not have done the perfect thing, but it did a safe thing,” Edwin Olson, co-founder and CEO of AV startup May Mobility, told Tech Brew during a recent ride-along in Ann Arbor, Michigan.
This is exactly the sort of somewhat tricky but all-too-common driving scenario that May Mobility’s proprietary technology, Multi-Policy Decision Making (MPDM), is designed to navigate.
The ride-along was fully driverless, meaning there was no safety driver behind the wheel, unlike a similar trip we took last year. It’s one of the latest signs of progress for a startup that is now positioning itself to become the sector’s leading autonomy-as-a-service provider.
“I think right now, May Mobility is in pole position to be the No. 1 provider of autonomy technology to other companies—the companies that need an autonomy play, and view Waymo as more of a threat,” he said.
Companies like Lyft and Uber, both of which have recently announced partnerships with May as part of their own strategies to get more AVs onto their ride-hailing networks.
May’s AVs will launch on Uber’s ride-hailing network later this year, starting in Arlington, Texas, where May has operated since 2021. In a recent press release, May said that it “aims to deploy thousands of AVs on the Uber platform over the next few years.” The companies said they plan to “expand to additional US markets in 2026.”
Keep up with the innovative tech transforming business
Tech Brew keeps business leaders up-to-date on the latest innovations, automation advances, policy shifts, and more, so they can make informed decisions about tech.
May AVs are slated to debut on Lyft’s ride-hailing platform in Atlanta later this year.
May also operates driver-free autonomous microtransit services in small, geofenced communities, like Peachtree Corners, Georgia, and Sun City, Arizona.
“I think a lot of people assume that you start off with a safety driver, and then you just record how often you needed to intervene, and when that number gets small enough, then you’re good,” Olson said. “But the problem is that that never happens. Driving is full of uncomfortable, risky scenarios.”
During our ride, a vehicle ahead of us stopped at an angle across both lanes, prompting the AV to stop.
“We almost certainly recognized that the car was in a strange orientation,” Olson explained. “And so when its behavior doesn’t match any of our models, we can basically flag it and be like, ‘WTF?’”
May’s system does not attempt to learn in advance every single possible scenario an AV might encounter.
“The way that most people think about autonomous driving is data-driven,” Olson said. “The AI successes over the last 20 years have basically been a triumph of collect a lot of data, train a model, profit. It’s reasonable to apply that same strategy to autonomous driving.”
The issue, he said, is that driving involves a basically infinite number of variables.
“The idea that you’re going to be able to rely on having seen a similar situation in your training data ahead of time doesn’t scale,” Olson said.
MPDM, meanwhile, uses “AI to interpret data in real-time, continuously learning and adapting to new, complex and even unpredictable driving conditions,” according to the company.
“You have to be able to reason through a situation you’ve never seen before,” Olson said. “This is what humans do.”