Delivery bot chief touts partial autonomy

Among purists blazing the path to a self-driving era, the idea of partial autonomy often induces cringes. Even among his own partners and investors, the phrase dampens enthusiasm, Felipe Chavez fears.

"Semi-autonomous" deserves more respect, he says.

Kiwibot, co-founded by CEO Chavez, launched 25 delivery robots — with a little help from remote human supervisors — in San Jose, Calif., last month. They can navigate sidewalks and street crossings almost entirely on their own. Once in a while, if they need help with a complex situation, a teleoperator assists.

Self-driving technology will come someday. In the meantime, automated technology that involves human assistance can allow for meaningful progress.

"I'm really a contrarian on this," Chavez said. "But we have an autonomous robot that can avoid obstacles. The corner-case scenarios, the part that's challenging, we have supervisors that can handle three robots at the same time. The good news? Our robots work today. We don't need to wait several years."

Such a development allows the company to deploy and generate revenue today, when demand for delivery services is booming. Kiwibot has already completed 110,000 orders in various pilot projects.

In San Jose, businesses using the bots are charged $3.99 per delivery. At that price, Chavez says, Kiwibot presents an attractive alternative to a delivery business model that irritates restaurateurs with high fees while couriers earn little and delivery companies operate on low margins.

"When I talk with customers, I'm very comfortable that I can beat what they're paying right now," he said.

Kiwibot is working with roughly 100 restaurants in San Jose on food delivery and is quickly adding more. In July, the company partnered with Shopify and Ordermark, allowing more businesses to integrate their delivery needs with the robots.

By the end of next year, Chavez expects to have 500 robots in operation in San Jose and beyond. Roughly the size of a cooler, the newest bots are designed to hold a 12-inch pizza. Their width is about the same as a human's, and they travel at 3 to 5 mph, about the same as a jogger. Those are all considerations made with the desire to mingle with — not frighten — humans.

"People think of it more as a pet," Chavez said. "They can't feel threatened."

That's important because the company's bots are deployed on sidewalks, in contrast to other delivery-focused companies that have deployed their vehicles on roads (Nuro) or in bike lanes (Refraction AI).

In October, experimental food delivery robots were pulled from University of Pittsburgh campus sidewalks after a backlash. To avoid a similar fate for Kiwibot, Chavez has been working closely with San Jose officials.

At first, the company was attracted to the Japantown neighborhood, but that community recently fought to get e-scooters removed from its sidewalks. Residents there might have been apprehensive about another vehicle, even one that moves at considerably lower speeds than scooters, said Andrea Arjona, project manager with the city. Services instead began downtown and in the city's Buena Vista neighborhood.

"We want to first demonstrate what happens in other areas of the city that are more amenable," she said. "And Kiwibot has been adaptable. Every suggestion we made to them — it didn't matter how crazy or difficult it was."

Kiwibot and San Jose have joined in a first-of-its-kind collaboration to feed location information from the delivery bots into the city's Mobility Data Specification, a tool cities use to collect data from passenger-carrying mobility modes, such as e-scooters, rental bikes and ride-hailing services.

From a city perspective, it's an important development that could be replicated elsewhere.

"After scooters, one thing we're trying to do is not get caught by surprise again by a new vehicle moving in," Arjona said. With the mobility data tool, "there's no nomenclature for robots, but we're adapting one. So we will have an update we can expand to new modes. That's pretty significant."

The collaboration presents potential benefits for citizens and robots alike. If an intersection is daunting for robots to cross, it is likely troublesome for pedestrians. If a crumbling sidewalk is difficult to traverse for bots, it likely is challenging for a wheelchair. Now such insights can be shared.

"We know which intersections are more dangerous, where drivers have more risky behavior," Chavez said. "This is very exciting for us, and it's very exciting for policymakers. It's allowing us to have better conversations with other cities."