Refraction AI

By A2Hub Editorial. Published 2026-04-22. Last updated: 2026-04-22.

At a glance

• Founded: 2019
• Founders: Matt Johnson-Roberson, Ram Vasudevan (both U-M faculty)
• HQ: Ann Arbor, Michigan
• Product: REV-1, a three-wheeled lightweight autonomous delivery robot designed for bike lanes
• Funding raised: ~$10.7M across three rounds
• Key investors: Pillar VC, Trucks VC, Alumni Ventures, Invest Michigan
• Status: Operational; Ann Arbor footprint scaled back per recent Ann Arbor Observer reporting
• Website: refraction.ai

The contrarian bet on bike lanes

Most autonomous delivery startups ran straight at one of two ends of the size spectrum. They either built tiny sidewalk robots in the Starship Technologies mold, slow and small enough to share a curb with pedestrians, or they built full passenger-class autonomous vehicles capable of mixing with car traffic. Refraction AI picked the gap nobody else wanted.

The REV-1 is roughly the size of an electric scooter cargo box. It rides in the bike lane at bicycle speeds, which puts it in a regulatory zone where most cities already permit micromobility. It is light enough that a collision is a property-damage event rather than a personal-injury event. It is heavy enough to carry real grocery loads, restaurant orders, and pharmacy deliveries. The technical claim that founders Matt Johnson-Roberson and Ram Vasudevan staked the company on is that the bike lane is the right operating domain for last-mile robotics, because it eliminates most of the failure modes that make sidewalk and roadway autonomy hard.

That bet is unfashionable. In a sector where every funded competitor is either a Bay Area sidewalk robot or a Phoenix-style passenger AV, Refraction AI built something specifically suited to the four-season weather, narrow streets, and bike-lane infrastructure of an upper-Midwest college town. Ann Arbor was the live testbed.

Founding lineage: U-M robotics

Matt Johnson-Roberson and Ram Vasudevan are both faculty at the University of Michigan. Johnson-Roberson came out of the U-M Department of Naval Architecture and Marine Engineering with deep experience in underwater autonomy, computer vision for unstructured environments, and SLAM in conditions where GPS does not exist. Vasudevan is a U-M mechanical engineering and robotics professor focused on safe motion planning for robots operating around humans.

Their joint background is unusual in autonomous delivery, where the typical founding team is a Bay Area robotics PhD plus an operations executive from a delivery company. The Refraction founding pair leaned harder into robotics fundamentals, particularly the safe-motion-planning research that lets a robot make conservative decisions when its environment goes sideways. That is the right team for a robot that has to share a bike lane with humans on e-bikes who are not paying attention.

The company spun out of U-M in 2019 and put its first robots on Ann Arbor streets within months. By late 2019 and into 2020, Refraction had partnerships with downtown Ann Arbor restaurants for last-mile food delivery. The pandemic pulled forward demand for contactless delivery and gave the company more real-world miles than its competitors managed in the same window.

Funding and growth

Refraction raised roughly $10.7M across three rounds. The seed was $4.2M led by Pillar VC, with Trucks VC, Alumni Ventures, and Invest Michigan participating. Subsequent rounds extended runway and funded fleet expansion. The company never raised at the SoftBank-style scale of the sidewalk-robot generation that came to market in 2017 and 2018, which has turned out to be a feature rather than a bug. Several of those better-funded competitors burned out or pivoted in 2023 and 2024 after the easy-money window closed.

Recent reporting in the Ann Arbor Observer indicates Refraction has scaled back its Ann Arbor street operations. The company's broader operational status, customer roster outside Ann Arbor, and current headcount are unconfirmed at the disclosure level. What is clear is that the technology and the IP remain. The team built one of the only small-format autonomous delivery vehicles purpose-engineered for cold weather and bike-lane operation, which is a thin and defensible market position if and when the regulatory and unit-economics picture stabilizes.

What they build

The REV-1 is a three-wheeled, electric-powered autonomous vehicle roughly five feet tall and three feet wide. It carries cargo in a temperature-controlled compartment sized for grocery bags or several restaurant orders. The vehicle is designed to operate in bike lanes at speeds up to roughly 15 mph, with a sensor stack tuned for the specific failure modes of bike-lane operation: cyclists overtaking on the left, pedestrians crossing unpredictably, parked cars opening doors, and weather conditions that degrade lidar and camera performance.

The autonomy stack reflects the founders' research backgrounds. Where many delivery-robot startups leaned on heavy lidar and high-definition maps, Refraction has emphasized lighter-weight sensor fusion combined with safe-motion-planning algorithms that default to slow or stopped behavior when uncertainty rises. The robot is engineered to fail safe rather than fail fast.

Why Ann Arbor

Refraction's Ann Arbor base is not incidental. The city offered three things the founders needed in close proximity: U-M robotics talent for hiring, downtown restaurant density for delivery demand, and a city government willing to permit autonomous bike-lane operation as part of the broader mobility-corridor experimentation that runs from Mcity through May Mobility's A2GO shuttle pilot to Refraction's robots.

The flip side of that local-laboratory advantage is the same constraint that has made Ann Arbor a hard market for several mobility startups: the city is small. Real commercial scale requires expanding into denser metros where bike-lane infrastructure is uneven and weather conditions vary. The execution path from "Ann Arbor pilot" to "10-city deployment" is the gap the company has been working through.

Related

• Quantum Signal AI, the Saline-based Ford subsidiary that builds simulation and autonomy software, is the other Washtenaw County mobility company with deep U-M roots.
• May Mobility is the Ann Arbor autonomous shuttle operator and the most-funded mobility company in the region.
• Ann Arbor mobility companies is the category hub covering every AV, robotics, and mobility firm in the area.
• Companies in Ann Arbor is the master directory.
• Ann Arbor tech hub is the homepage pillar on the broader ecosystem.

FAQ

Who founded Refraction AI? Matt Johnson-Roberson and Ram Vasudevan, both faculty at the University of Michigan, founded Refraction AI in 2019. Johnson-Roberson's research background is in underwater autonomy and computer vision; Vasudevan's is in safe motion planning for robots operating around humans.

What is the REV-1? The REV-1 is Refraction AI's three-wheeled autonomous delivery robot. It is designed to operate in bike lanes at speeds up to roughly 15 mph, carrying groceries, restaurant orders, and pharmacy deliveries in a temperature-controlled compartment.

How much has Refraction AI raised? Refraction AI has raised approximately $10.7M across three rounds, including a $4.2M seed led by Pillar VC with participation from Trucks VC, Alumni Ventures, and Invest Michigan.

Is Refraction AI still operating in Ann Arbor? Per recent reporting in the Ann Arbor Observer, Refraction has scaled back its Ann Arbor street operations. The company's current operational footprint and headcount are not publicly disclosed.

Why did Refraction choose bike lanes instead of sidewalks? The founders' bet is that bike lanes are the right operating domain for last-mile delivery robots because they eliminate sidewalk-pedestrian conflict, give the robot legal access to higher operating speeds, and use micromobility regulations that are already in place in most US cities.