Memo may not be the world’s fastest barista, but it is impressive—for a robot.
I recently watched as Memo, a new home robot from a company called Sunday Robotics, made coffee in an open-plan kitchen in Mountain View, California.
Memo looks like something out of Wall-E, with a gleaming white body, two arms, a friendly cartoonish face, and a red baseball cap. Rather than using legs as a fully humanoid robot would, Memo moves around using a wheeled platform and changes its height by sliding up and down a central column atop that platform.
The robot responded to a request for an espresso by rolling over to a countertop, and then using two pincerlike hands to slowly go through each step required to operate an espresso machine. It filled the porta filter with coffee grounds, tamped them down, slotted the porta filter into place and put a coffee cup below, pressed the buttons needed to start the machine, and finally retrieved the hot drink.
“We want to build robots that free people from laundry, from the dishes, from all chores,” Tony Zhao, cofounder and CEO of Sunday Robotics, told me as the robot brought the coffee over to the person who requested it.
Making a cup of espresso might not seem spectacular, but the feat is ridiculously hard for a robot to do in a real, messy kitchen. It requires the ability to identify different objects, figure out how to grasp them reliably, and use those objects properly. Sunday is not only building its own hardware but also training the models that allow its system to learn. “We think the way to make a home robot is to be full-stack, and to vertically integrate,” Zhao says. “And that’s a very ambitious thing to do.”
Courtesy of Sunday Robotics
Today most robots still do precise, repetitive work in tightly controlled environments; for example, moving the same item from one position to another over and over again. Unlike humans, industrial robots cannot typically adapt or improvise to changes or unfamiliar situations. The last decade has seen some companies build robots that use AI to do simple things, like identify objects on a conveyor belt and decide how to grasp them. This is, however, far less complex than operating in an environment as varied and messy as a real home.
Of course, robot demos are not always a good indicator of how useful a robot will be. The real question is how well Memo can perform tasks in a wide variety of homes without Sunday’s engineers nearby.
Even so, the robot has some skills. Besides making coffee, I watched Memo clear glasses from a table and load them into a dishwasher. This feat was especially impressive because it involved figuring out how to grasp two glasses in the same hand. Memo held one glass between its thumb and pointer finger, and used the rest of its hand to grab the second.
This kind of dexterity relies on Sunday’s key innovation: a novel way of training robots that delivers more humanlike dexterity. Sunday pays remote workers to use gloves resembling Memo’s hands to do household chores. Zhao says the gloves, which cost roughly $400 a pair, provide a more accurate training signal than teleoperation, which is the standard way for a person to control a robot. The training data gathered from glove-wearing workers is fed into an AI model that takes input from the robot’s sensors and controls its motions.
“This is a very exciting variant on home robots,” says Ken Goldberg, a roboticist at UC Berkeley and the cofounder of Ambi Robotics. “It’s a beautiful design, and a much smarter kind of data capture.”
Courtesy of Sunday Robotics
The fact that any company thinks it can build a useful and functional home robot is a sign of skyrocketing optimism about progress in robotics—and not without reason. Researchers have shown in the last few years that robots can tap into the capabilities of large language models, the brains behind today’s chatbots, and use them to respond to commands or make sense of a scene.
Some researchers hope that gathering large amounts of data that shows how to perform different actions—picking up cups, folding shirts, and so on—will produce a more general kind of robotic intelligence.
Zhao and Sunday’s other cofounder and CTO, Cheng Chi, have both contributed advances that have kindled hope of robotic breakthroughs. Zhao worked on a project called Mobile ALOHA at Stanford University that involved training robots using a low-cost mobile teleoperation system. Chi worked on a project from Stanford, Columbia University, and the Toyota Research Institute, that showed how a cheap clawlike device could be used to gather data from humans doing tasks like cleaning dishes.
“If you think about the most powerful AIs, ChatGPT or image-generation models,” Zhao says, “they are trained on the whole internet. We just don’t have the internet for robotics.”
A handful of other startups are currently hustling to develop and deploy more capable robots, including systems designed to work in ordinary homes. Physical Intelligence, Skild, and Generalist are all working on robot models that can adapt to new situations using this approach. 1x recently revealed a humanoid home robot, though this system still requires teleoperation to perform some tasks.
Sarah Guo, founder and managing partner of Conviction, says that Sunday, which includes veterans from Tesla and Google DeepMind, is well placed to build both hardware and models. “Tony and Cheng are incredibly good,” she says. “They’ve since recruited and enabled an all-star team that can uniquely do hardware–AI co-design, and deliver a full-stack product.”
Eric Vishria, a general partner at venture capital firm Benchmark, which is backing Sunday, said in a statement that the startup’s practical approach is the way to make robots more useful. “The promise of AI robotics isn’t doing a backflip or dancing demos, but robots that work in messy, real-world situations,” Vishria said, adding that Sunday’s “breakthroughs mark the start of an exponential curve toward a future where robots actually work in our day-to-day lives.”
Sunday plans to give Memo to beta testers next year. The pilot program will show how people respond to having a home robot that can do certain chores—albeit slowly, and perhaps not perfectly every time. A key question will be how reliably and efficiently Memo is able to do chores in real homes where kids, pets, and mess are guaranteed to complicate the challenge.
After beta testing, Zhao says Sunday will roll Memo out to the first users. Just as early home computers were complicated and appealed mostly to enthusiasts, he believes Memo might initially be popular with those who want to live in a robotic future and are willing to tolerate some rough edges. This might even involve users showing their robots how to do something new. “I do think that people should be able to teach their own robots,” Zhao says.
Perhaps the era of truly capable home robots is almost upon us. For now though, I’d settle for a decent espresso.
Updated 12:23 pm ET, November 19, 2025 to clarify the cost of the gloves used for collecting robotic training data.
Updated 1:26 pm ET, November 19, 2025: Added comments from Sarah Guo and additional details about the Sunday team.
