Matthew Johnson-Roberson grew tired of waiting.
He had been part of the self-driving industry since its nascent days. Then he watched his hopes of seeing autonomous vehicles in widespread deployment get dashed as a laundry list of projects were delayed or simply went away.
When he started his own company, Refraction AI, in 2017, he intended to use autonomous tech to make a near-term impact. So he and co-founder Ram Vasudevan focused on delivering food instead of carrying human passengers. During a pandemic that's heightened interest in delivery services, he's getting that chance.
Refraction AI has launched a pilot project in Ann Arbor, Mich., using eight of its prototype three-wheeled robots to deliver goods from a local grocer. The testing builds upon a similar project started in December, in which the vehicles deliver meals from several local restaurants.
"It was frustrating to see full-sized AVs not coming as soon as we'd hoped, and I wanted to put something out on the road now, safely," said Johnson-Roberson, who worked on Carnegie Mellon's team, under robotics pioneers Red Whittaker and Chris Urmson, during a landmark 2004 AV competition staged by the Defense Advanced Research Projects Agency.
"So we've been building something that we can put out there on the roads and use to conceivably take over big chunks of the last-mile delivery market, and all that was pre-COVID," said Johnson-Roberson, who works as a robotics professor at the University of Michigan in addition to his CEO duties at Refraction AI. "Then the crisis has really brought into sharper focus the need for contactless delivery."
Since the pandemic struck, Johnson-Roberson says the number of trips taken by the vehicles has quadrupled. Working with Roush, which builds Refraction's custom prototypes, the company hopes to have 18 more of its vehicles deployed in the Ann Arbor area by the summer.
Consumer demand for delivery services has mushroomed during the pandemic. The World Economic Forum says that with the rising popularity of e-commerce, the number of delivery vehicles in urban areas will jump 36 percent by 2030.
Developers have been keen to explore the role of self-driving tech in that delivery space, especially now.
"This pandemic has brought the need for automated delivery solutions to a new level," said Asad Hussain, mobility analyst at PitchBook. "Before, investors and management teams were interested in the autonomous delivery space as a means to reduce delivery costs. Now it's about ensuring safety for consumers and continuity of operations in a rapidly expanding market with an emphasis on contactless delivery."
For Refraction's founders, smaller delivery vehicles seemed like an ideal use case for self-driving technology, and they came without the baggage of human expectations in either the driving or riding experience. For example, a route that included four right turns to avoid a complicated intersection might irk a human but doesn't matter to a bag of vegetables.
Where these bots should travel on public infrastructure remains an open question, and developers are creating vehicles for a range of environments. Some companies have developed grocery-cart-sized robots strictly for sidewalk use. Nuro, on the other hand, built its larger R2 vehicles specifically for on-road use.
Refraction AI believes there's a sweet spot between the two. Its REV-1 prototypes are tailored for travel in bike lanes. Traveling at 12 to 15 mph and weighing about 100 pounds, the delivery vehicles are most akin to a bike with a teenage rider. With three wheels and a small electric motor, they fit legal definitions of an e-bike, which Johnson-Roberson says is advantageous for regulatory and insurance reasons.
The REV-1 operates with camera, radar and ultrasonic sensors, and has a stopping distance of about 5 feet, which Johnson-Roberson says brings added safety compared with competitors.
At that speed and stopping distance, Johnson-Roberson says, "the worst thing that happens is your burrito gets smashed."
A vehicle can travel roughly 150 miles per battery charge, which amounts to eight to 10 deliveries per charge. Battery packs can be easily swapped. A remote operator — working from home during the pandemic — has been watching the vehicles, but notably, Johnson-Roberson says there are no chase vehicles in the field keeping an eye on the fleet.
For now, Refraction AI is focused on Ann Arbor. Eventually, the company hopes to expand to Grand Rapids, Mich., and Detroit. Much like the company's vehicles attempting to find a middle ground between the sidewalk and road, Refraction AI's business blueprint calls for finding sweet spots between cities too small to support frequent deliveries and those so large that city environments are too chaotic for operations.
"We need density, but we also need it to not be too crazy," said Johnson-Roberson, who said the ideal density is about 4,000 people per square mile. "That feels like the right number for the moment. … So the next few years we're targeting places with a lot of density."