Ann Arbor restaurant and food-market proprietors and their customers have been delighted with their robotic delivery REV-1 bots designed, engineered and built by Refraction AI. (Refraction AI)
Michigan start-up Refraction AI aims to be the go-to for last-mile, semi-autonomous, all-weather delivery robots.
When road users and pedestrians in downtown Ann Arbor, Michigan, first view the REV-1 in action, they see a funky little electric three-wheeler quietly navigating the city’s streets on its own. Observers marvel when the vehicle obeys traffic signals, stops at crosswalks, deftly avoids parked cars and gracefully shares bike lanes. Not readily apparent beyond REV-1’s unique form factor are its surprising 3.5-mile (5.6-km) operating radius and workhorse 7-hour duty cycle. Both seem increasingly right for app-driven, socially-distanced local commerce.
In combination with a 12-camera array, a radar, and ultrasonic sensors, REV-1 uses an unnamed low-cost lidar. Engineers at Refraction AI, which created the vehicle, see it as a game-changer for smaller, less-costly autonomous vehicles. Many of these innovative systems are aiming for a chunk of the “last mile” urban delivery market that is widely forecast to yield compound annual growth rates above 12% by 2025.
REV is a lightweight, compact dynamo – 54 inches long, 30 inches wide, and 48 inches tall (1371 x 762 x 1219 mm). Using an air-cooled lithium-ion battery for its onboard energy source and shaft final drive, the trike weighs approximately 100 lb (45 kg) and can achieve velocities up to 15 mph (24 km/h). REV’s 16-ft3 (453-L) cargo compartment holds up to seven bags of groceries or the equivalent payload of take-out restaurant meals, or packages. A heating element is under development for hot items such as pizza.
“We can handle a variety of tasks, we’re not just restricted to food deliveries,” explained Ramanarayan “Ram” Vasudevan, a co-founder of Refraction AI. Vasudevan is an assistant professor of Mech. Eng. at the University of Michigan, who has developed walking robots and manipulators. His co-founder colleague Matt Johnson-Roberson also teaches robotics at U-M and is a specialist in underwater autonomous systems. They started the company in 2019 with assistance from the university’s Office of Technology Transfer, along with backing from Trucks Venture Capital and eLab Ventures.
Refraction AI currently is operating eight REV-1 delivery ‘bots in a pilot development program, aligned with a like number of Ann Arbor restaurants and grocery stores. Customers living within the delivery area place their orders with these businesses through a custom link. They then receive a coded text message with delivery updates – the goal is to have food and parcels in hand within 40 minutes, Vasudevan said. Customers are alerted to meet REV-1 at the curb; when it arrives, they input the keypad code. This opens the ‘bot’s cargo door, revealing their order. The service, aimed at both proving out REV-1 and acclimating local businesses to the ADV (autonomous delivery vehicle) business model, is currently free of charge.
ADVs “will dramatically drive down operations costs, making dense networks less essential and opening the door to smaller, newer players,” noted McKinsey & Co. analysts Jürgen Schröder and Bernd Heid in their recent study: Fast forwarding last-mile delivery – implications for the ecosystem.
Developing to a low price point
REV-1 and the idea of Refraction AI were born from the classic napkin-sketched brainstorming engineers know well. “Autonomous vehicles have the promise of really changing society,” Vasudevan said. “The challenge we saw is getting them operating reliably and safely, in the inclement weather and difficult road conditions that most of the country lives in. So, we wanted to bring autonomous systems out into the real world, testing the vehicle in a true four-season environment. That’s a big challenge that we’re facing head-on.”
The Refraction AI team did its pilot testing in the Michigan winter, which Vasudevan deems “vital.” One of his ongoing disappointments had been that so much first-generation AV testing was being done in moderate-climate California. “We made it our mission to bring it here,” he said.
The co-founders decided on a low-speed, low-cost platform operating on the margin of the road; primarily in bike lanes, not on the sidewalk as some other systems do, where they conflict with pedestrians. “What we saw, looking at our wealth of experience, is this [last-mile urban delivery robots] is a ‘sweet spot’ with respect to various design requirements that our platform fulfils,” Vasudevan said. “We’re primarily focused on developing a technology, a tool, that can be applied today – and proving out a business model that makes sense.”
That means keeping the price point of REV-1 and its successors affordable for both Refraction AI and for its customers – the entire sensor platform costs a fraction of a single HD lidar, the company maintains. “We need to ensure every partner who sits down with us is actually walking away with money in their pocket and feeling good about the business we’re both working on.” Some businesses were skeptical on first approach by Refraction AI; others were “on board immediately,” he noted.
A strategic goal of the company’s 12 engineers and technical staff was to keep the REV-1 relatively simple and low cost. That meant rationalizing the sensor suite. Lidar was viewed necessary in “simplifying some of the autonomous challenges,” Vasudevan explained, but a full-capability large HD lidar was too expensive: a Velodyne 128, for example, costs a few hundred thousand dollars. That doesn’t quite fit Refraction AI’s business model or the size and weight of its platform.
A less-costly lidar sensor is providing “some of the attractive features of that full-scale lidar, perhaps without the full-scale range. Figuring out ways to use alternative sensing modalities and other tools to address those challenges well has been fun,” Vasudevan explained without detail. “The good news is our slow road speed enables our cameras and the rest of our sensor suite to fill in that gap really well. Some of that comes from us working with suppliers who have novel sensors and who are trying to figure out a way to market them – but haven’t found the right use case for them.”
Thermal sensors – microbolometers that detect infrared radiation with wavelengths between 7.5–14 μm – have been investigated for REV-1 testing. But their challenge is excessive “motion blur”’ associated with them, along with cost. The Refraction AI co-founders’ robotics students have been working on techniques to employ “very inexpensive” thermal cameras, “but they still need a lot of work before they can be deployed.”
REV-1 also is tele-operated – controlled by a combination of artificial intelligence and human operators working via wireless remote. “Our operators watch the robots from their homes right now,” Vasudevan said. “There will always be situations with autonomy, at least in the near term, that a human will be required to help the robot or vehicle through the most complicated parts of the delivery situation.”
To optimize uptime and simplify maintenance, the REV-1 platform is designed like a funny-car dragster, with a body that pivots up to service the subsystems. “If we need to change a battery we can hot-swap it in quickly,” Vasudevan said.
More REVs from Roush
Increasing demand for REV-1 services have prompted Refraction AI and its investors to contract Roush Industries to help expand production with the aim of adding 25 new delivery ‘bots this fall. Building an effective engineering team at a start-up is easier when the company’s two principals are academics who can recruit high-pot talent from their robotics and AV classes. “The students who we’ve been able to bring over to the company have been great engineers for us,” Vasudevan said. But during the COVID lockdown, maintaining product-development schedules, meeting deadlines, managing iteration cycles and staying productive was tough, he explained.
“It’s non-trivial to figure out how to do vehicle autonomy well. There are engineering tasks that are team-driven, where you need to have an engineer standing next to you or nearby,” he said. “Helping each other do things is really important, and our team has been really quick on its feet.” The Refraction AI team devised “a number of clever ideas” to manage design engineering in a way where space can be shared without people ganging together.
They also switched its remote-teleoperator system to home locations. More recently, as people have migrated back to work, working in shifts has proven effective. Within five years, Refraction’s goal is to have its robots doing urban deliveries of all types throughout the U.S. “If mail or other types of deliveries appear, we can be the go-to platform,” Vasudevan said.
By
Source: https://www.sae.org
CUT COTS OF THE FLEET WITH OUR AUDIT PROGRAM
The audit is a key tool to know the overall status and provide the analysis, the assessment, the advice, the suggestions and the actions to take in order to cut costs and increase the efficiency and efficacy of the fleet. We propose the following fleet management audit.
