Refraction AI’s 4.5-foot tall, 150-pound Rev-1 three-wheeled robot vehicle has been delivering pizzas in Austin, Texas. The company believes the future of autonomous driving is ‘zero occupant.’
Source: Refraction AI
As companies like Tesla and its CEO Elon Musk come to Austin, Texas, the booming city and new tech hub has grown so much it has struggled to make good on its “keep it weird” motto.
But since early June, when residents of the South Congress, Downtown, or Travis Heights neighborhoods order pizza from Southside Flying Pizza, their pies might arrive inside a three-wheeled robot — the REV-1. But it is no full self-driving Tesla.
About two dozen REV-1 vehicles now trundle down the roads of Austin and Ann Arbor, Michigan, where the company behind the robots — Refraction AI — first launched in 2019 in a bid to harness driverless technology in a new way.
Autonomous vehicles, and their potential to disrupt the way people get around, have hovered on the horizon for years. But the technology hasn’t matured as dramatically as early investors had hoped. Tesla says it is full speed ahead with autonomous and is launching its latest beta test on July 10, but the company has missed many self-imposed deadlines and Musk recently conceded that full self-driving is harder than he had forecast. Meanwhile, both Uber and Lyft sold off their self-driving research divisions in recent months.
For start-ups like Refraction AI, though, the aim is to speed the arrival of a driverless future by starting small, with more modest systems focused on moving packages, rather than people.
“We think this is the future of autonomy,” Refraction AI CEO Luke Schneider said.
While one might technically call the electric REV-1 a driverless vehicle, it’s a completely different beast from the LiDAR-equipped multi-ton vans in development from Alphabet-backed Waymo, General Motors-backed Cruise, and other multibillion-dollar ventures. Musk has bashed LiDAR in the past, calling it a “crutch,” though recent reports indicate the company has recently used the technology for reasons that remain unclear.
The REV-1 “trikes” stand 4.5 feet tall and weigh 150 pounds, making their profile more like bicycle couriers than delivery vans. And they act like cyclists too. After a restaurant worker places the “payload” in a REV-1’s storage compartment (which fits about six grocery bags), the robot uses an array of sensors to navigate to the edge of the road, or, if available, a bike lane. It then drives away at no more than 15 miles per hour toward its destination, where the customer greets it at the curb and unlocks their meal or package with a unique code.
If the vehicle runs into trouble along the way, which could include unusual obstacles like a curbed couch, or common but tricky-to-automate moves like turning left across traffic or navigating crosswalks, a dozen “pilots” are standing by to remotely and temporarily take control of the REV-1.
This is going to take a lot more money and a lot more time than we thought it was. It’s not going to be a flip switch overnight.
Luke Schneider, Refraction AI CEO
In Schneider’s eyes, these distinctions give the REV-1 a crucial edge in the delivery business. The small, slow carts can’t do much damage in the event of a malfunction. They don’t hog the road. And the outsourcing of the trickiest driving to the pilots means the company is ready to deliver food today. Spending years and billions of dollars to develop a fully autonomous 4,000-pound vehicle and then using it to deliver pizzas, to Schneider, is overkill. “I call it bringing a ballistic missile to a knife fight,” he says.
So far, the company appears to be making progress in solving this facet of the overall challenge of autonomous driving. They recently reconfigured the REV-1’s sensors to give it night vision, allowing it to work during the most popular time for meal delivery. Schneider says the REV-1 fleet carried seven times more meals early this year than it did late last, and has continued to make more deliveries every month since March. The 50-employee company recently raised 4.2 million dollars in seed money, for a total funding of $8 million.
Refraction AI isn’t the only self-driving start-up thinking small.
Nuro, a California company that has raised about $1.5 billion in funding according to Crunchbase, has developed a golf-cart sized delivery vehicle called the R2 capable of driving at 25 miles per hour. In April, the R2 began delivering pizzas for Domino’s Pizza in Houston, Texas. Last week, the company announced that its next-generation vehicle will deliver packages in a partnership with FedEx.
Domino’s tests Nuro, an autonomous car for pizza delivery in Houston.
Nuro was founded by Dave Ferguson and Jiajun Zhu, two former Alphabet engineers whose self-driving enterprise became Waymo. They were motivated by an assumption similar to that driving Refraction AI: Getting people entirely out of the equation would allow for a lower stakes first application of driverless vehicles.
“We set out to build a new class of vehicle, designed purely for transporting things,” wrote Ferguson in a 2020 Medium post. “A zero-occupant vehicle.”
A Chinese startup, Unity Drive Innovation, has taken a similar approach to distributing vegetables and meal boxes during the pandemic.
Even as mini-delivery carts start to venture out into residential streets and bike lanes, self-driving behemoths continue to doggedly pursue the technology’s full potential: vehicles that can carry anyone or anything anywhere without a driver.
Alphabet’s Waymo initially rolled out a partially autonomous beta test of its ride hailing service, Waymo One, in Phoenix, Arizona, in 2018. Last October, it opened the service to the public, simultaneously making it fully autonomous, with no drivers and no remote pilots — although it does use a “fleet response team,” specialists who feed the cars information when the machines can’t interpret an ambiguous situation, such as construction or road closures.
Waymo has also worked with UPS in Phoenix to shuttle packages between UPS stores and a local shipping hub. And two weeks ago it announced a partnership with trucking company J.B. Hunt to autonomously haul cargo across a route in Texas, initially under human supervision.
The company’s leadership remains confident in what Julianne McGoldrick, a Waymo spokeswoman, describes as a flywheel effect. Once you develop a vehicle that can navigate the roads about as well as a human can, you can use it for any number of applications that will mutually reinforce one other.
“We have seen that all the progress we’re making on ride hailing and on passenger cars directly feed into the trucking and local delivery areas, and then the progress we make there also goes back into ride hailing,” McGoldrick says.
Waymo, Nuro and Refraction AI are proving that machines can navigate the world well enough to start ferrying around people and pizzas in small numbers.
But the billion-dollar question remains: Will either model prove profitable enough to make self-driving vehicles go mainstream, potentially reducing the world’s carbon emissions and auto-accidents?
Ashley Nunes, a transportation research fellow at Harvard Law School, says that driverless vehicles might supplement current ride hailing and delivery services in places like densely packed urban areas with mild weather. But he suspects the harsh economics of competition with personal vehicle ownership will challenge the truly transformational benefits. He points out that all self-driving fleets require oversight, whether in the form of remote pilots or a fleet response team, and that the price of this human labor limits how affordable the vehicles can become.
“‘Autonomous’ or ‘driverless’ does not mean ‘humanless,'” he says.
But that won’t stop companies like Nuro and Refraction AI from aiming to bring a future brimming with driverless vehicles closer, one pizza at a time.
“This is going to take a lot more money and a lot more time than we thought it was,” Schneider says. “It’s not going to be a flip switch overnight.”