Serve Robotics launches new business to become the AWS for robots

Serve Robotics, a shaper of self-driving sidewalk transportation robots, is expanding beyond its autonomous transportation operations successful a bid to go a exertion supplier to different companies gathering robots.

The company, which spun out of Uber’s acquisition of Postmates successful 2021 and went nationalist in 2024, said Thursday it volition motorboat a caller concern enactment called Physical AI Services.

Specifically, Serve Robotics volition connection customers its connectivity and information tools. Bedrock Robotics, a startup processing self-driving systems for operation equipment, is simply a customer, according to Serve. It’s besides moving with T-Mobile to bring AI computing powerfulness person to robots, Serve announced alongside its third-quarter net connected Thursday.

“We are solving the occupation of robots successful quality spaces,” Serve Robotics CEO and co-founder Ai Kashani told TechCrunch.

Serve Robotics already makes 2 kinds of robots, the cute googly-eyed sidewalk transportation robot and 1 designed for hospitals that it got done its acquisition of Diligent Robotics.

Those businesses are not going away, Kashani emphasized successful a caller interview. Instead, helium said determination is accidental to merchantability the exertion it developed for its ain robots to different companies processing robots for farms, warehouses, and different uncontrolled environments.

“I benignant of deliberation of it arsenic the AWS infinitesimal wherever we are taking the exertion we’ve built for us, and we’re opening it up for much radical to physique on,” helium said, adding this wouldn’t person been imaginable without processing its ain robot fleet.

The enlargement comes arsenic Serve’s sidewalk transportation concern expands, but is shifting distant from Uber and towards different partners, including DoorDash and Wonder.

Its caller Physical AI Services concern enactment volition connection 3 products. Serve Link helps link robots successful the tract to AI systems moving successful information centers, which would let companies to perchance pat into much almighty AI models without having to enactment each that computing powerfulness onboard.

“Any robotic exertion mostly needs to link the robots to immoderate offline intelligence, whether it’s to cod information oregon bash objection handling oregon whatever,” said Kashani. “It’s the connectivity portion that has to beryllium truly precocious performance. You request to beryllium capable to negociate truly unreliable networks and inactive reliably connect, and that’s what we solve.”

It’s besides launching a information curator instrumentality that helps companies find borderline cases successful the reams of real-world information a robot collects, which tin past beryllium fed backmost into a grooming exemplary to amended performance. Its 3rd merchandise is simply a real-world, foundation-model information catalog built from Serve’s ain sidewalk robots that tin thatch AI models however to recognize analyzable quality environments.

Kashani said he’s been reasoning astir this thought — a playbook helium learned from Amazon — for years.

“When the exertion is truthful nascent, you person got to physique the end-to-end exertion due to the fact that that’s the lone mode you cognize what to build, but past implicit time, arsenic the exertion gets much advanced, a batch much radical are going to beryllium wanting to physique akin things,” helium said. “And what you privation to bash is to unfastened your exertion truthful others physique with you and connected your platform.”