In cities like Houston, Phoenix and San Francisco, drivers have been sharing busy roads with autonomous delivery vehicles for months now.

In Ann Arbor, Mich., the creators of a new autonomous vehicle have designed their robot to operate on local streets — but more like a bicycle than a car. Built by a startup called “Refraction AI,” the REV-1 — a four-foot tall robot that weighs about 80 pounds and travels at a top speed of 15 mph — can operate in both car and bike lanes.

The company claims the three-wheeled vehicle’s ability to move between bike and automobile lanes gives the vehicle more flexibility and helps it avoid clogging traffic. It’s small size allows it to come to a full stop in about five feet, allowing the company to use lower cost sensors like cameras to see ahead and avoid accidents.

“We’re trying to emulate what it is to be a cyclist,” Matt Johnson-Roberson, the company’s co-founder and co-director of the UM Ford Center for Autonomous Vehicles (FCAV), told Wired.

The REV-1 is already delivering food from two Ann Arbor restaurants and making deliveries to the company’s employees. In the coming months, the company plans to expand its delivery service to the general public. Once service expands, Refraction AI will find itself competing in an increasingly busy space alongside big name companies like Kroger, Domino’s Pizza, Uber , Nuro, Waymo, Walmart, Amazon, Doordash and Starship Technologies that are vying for territory and experimenting with new ways to deliver food to customers.

With 16 cubic feet of space inside (enough to hold about four grocery bags), the REV-1 is one of the smallest and slowest autonomous delivery vehicles on the roads. By comparison, Nuro’s R1 autonomous vehicle — which has been delivering groceries for Kroger in Houston and Scottsdale, Ariz., since last year — can reach 25 mph and is about half the width of a Toyota Corolla.

Because of the vehicle’s larger size and ability to reach higher speeds, engineers designed it to “prioritize the safety of humans, other road users, and occupied vehicles over its contents," according to the company.

Unlike some of it’s competition, the REV-1 doesn’t rely on an expensive and complicated suite of navigational tools like lidar sensors. Instead, the vehicles rely primarily on multiple cameras, radar and ultrasound, allowing them to reduce the cost of the vehicle to around $5,000.

For now, the REV-1 is making deliveries between .5 and 2.5 miles away, a distance that could allow Refraction AI to tap into a market of customers eager for safe and quick last-mile delivery solutions, Bob Stefanski, managing director of eLab Ventures, told TechCrunch.

“Their vehicles are also lightweight enough to deploy more safely than a self-driving car or large robot,” Stefanski said. “The market is huge, especially in densely populated areas.”