The intersection at State Street and South University Avenue is one of the trickiest in Ann Arbor, Michigan. The three-way stop in the heart of the University of Michigan’s campus features a bus stop, students crossing in every direction with reckless abandon, and general automotive confusion over the right of way. Vexing to attentive drivers, it would seem a particularly daunting challenge for an autonomous vehicle like, say, the self-driving Lexus we were sitting in.
That Lexus RX450h is one of five May Mobility autonomous vehicle shuttles currently operating in Ann Arbor as part of a pilot program that launched October 11. After letting a pedestrian cross, the shuttle inched forward, sensing it was our turn to go. Just as we entered the intersection, a car shot forward to make a left turn, cutting in front of us and catching us by surprise. The shuttle spotted the aggressive move, hit the brakes, and avoided a potential crash before anyone could even gasp. Not bad.
The autonomous shuttle service is called A2GO, and it’s a collaboration with Mcity, a University of Michigan autonomous vehicle test facility, and SPARK, a nonprofit that supports economic development in Southeast Michigan. The service covers some of Ann Arbor’s densest areas while also linking transit hubs such as Blake Transit Center, where Ann Arbor’s buses depart, and the Amtrak train station. Shuttles, available at 18 designated stops, can be hailed for free via an app.
Lexus
May Mobility’s autonomous driving system, dubbed Multi-Policy Decision-Making (MPDM), is the brains behind the A2GO vehicles. It differs from many other AV systems in that it’s not strictly rule-based, where the car makes decisions based on a predetermined set of rules. Rule-based systems may seem intuitive, but AVs consistently encounter new situations that require additional, and sometimes conflicting, rules. “They become cumbersome, and every time you fix one thing, you break four other things,” says Edwin Olson, May Mobility’s CEO and co-founder. Adding more rules increases both complexity and cost, so May chose a different route.
“We use a simulator on the vehicle, running about 30,000 times faster than real time, to build a mirror copy of the world online in the car,” Olson said. The system continuously assesses the surrounding environment and reviews thousands of simulated decisions to determine what to do. As field engineer Jay Miles explained, MPDM is interaction aware, with the system simulating “what’s most likely to happen [with other road users], but also what they will do because of what other people are doing.” In a sense, it “gives the car an imagination and lets it work out on its own what decision makes sense given the context,” Olson added.
“If you use the analogy of a chess game, a chess grandmaster will look at the board and see all possible moves and take each one of those out several steps,” said Sam Abuelsamid, an AV expert and principal research analyst with Guidehouse Insights. “That’s kind of what they’re doing, and that requires both high-powered computing and very efficient software to run all of those scenarios simultaneously in real time.”
Lexus
A2GO is May Mobility’s eighth autonomous vehicle deployment, with the company previously serving cities from Arlington, Texas, to Hiroshima, Japan. Each city also has its own quirks. Whereas May found that Japanese drivers and pedestrians are predictable rule followers, “college students rule Ann Arbor” and “don’t really yield to let cars go through crosswalks,” Miles said.
But the system handled everything thrown its way on our three rides around Ann Arbor. That included brazen students, crumbling roads, and narrow streets. A2GO vehicles prioritize safety first, then rider comfort, and then autonomy. The system gets cautious when something sticks out into the road, like a dumpster or a parked truck, slowing to a crawl since it doesn’t want to cross double yellow lines.
While the vehicle could creep past the dumpster on its own, May’s has AV operators in each car trained to take over in such scenarios in order to provide a smoother, more normal experience for riders. “If the rider isn’t happy, then this vision isn’t going to move forward,” Olson said. “You give them a good experience and people will adopt it.”
Three rides is a small sample size, but the A2GO’s AV shuttles seem capable and safe; if you were blindfolded, you might only guess a human wasn’t steering when it exercises extra caution. The AV accelerates and stops smoothly—except in an emergency, when it really jams on the brakes. It also seems to be making a good impression on locals. On one ride we were joined by an engineering student from the University of Michigan who was enthusiastic about the technology and pleased with the experience, and all the operators we met—from a former professor to a self-proclaimed “car geek” who spent decades working in the auto industry—seemed genuinely impressed by the AV’s capabilities and potential to change the way we commute.
Still, self-driving cars have a long way to go. Olson reiterated that A2GO is a learning opportunity and that much more testing is needed before AVs become commonplace. But this shuttle also shows just how good the technology already is, and it proves the potential benefits of autonomous vehicles for city life. As Olson says, “You win by creating public transit that’s so damn good that people who can afford not to use it, use it anyway.”
This content is created and maintained by a third party, and imported onto this page to help users provide their email addresses. You may be able to find more information about this and similar content at piano.io
Source: www.caranddriver.com