Sunday Robotics

The Helpful Robotics Company

The first real project I ever built with code was a robot dog. Despite this, for an entire decade, I consistently told my investing partners: “Robotics is a great way to lose a lot of money over many years.”

In the basement of Stanford’s Robotics building, Tony Zhao and Cheng Chi changed my mind. I think they will help change the world. Sunday is building general-purpose, autonomous consumer home robots. Nothing speaks like shipping – their first robot Memo, and their model, ACT-1, are state of the art in dexterity and embodiment, and in scene and object generalization.

We are at a unique moment in robotics

For years, we’ve seen gradual improvements: more capable algorithms, cheaper and better sensors, motors, and more compute. But now we’re on the cusp of a step-change, where AI models can take robotics from brittle, single-purpose systems to autonomous, general-purpose machines.

Where should we begin?

There’s one obvious answer most people have shied away from because it’s too bright to look at: consumer home robots. Done right, they will be the greatest consumer product ever created. A general robot isn’t a single-purpose appliance (as robots have essentially been to date). It’s the next iPhone, a platform where every mundane task, chore, errand becomes a potential capability delivered through a software update.

The goal is simple and profound: free humans from daily toil so they can spend more time on their relationships and passions—the things that make us human.

Anyone who can afford one will buy one. We can design them so they cost closer to a smartphone than a car, and are constantly being more capable. They will change the human experience.

The core bottleneck: robot data

What stands between us and that future?

Robotics is still in a small-data age. Unlike LLMs, we do not have internet-scale data for embodiment. Scaling compute without data is not just inefficient; it won’t get us to truly capable home robots.

Video pre-training may get us some of the way. It can give us physical common sense like what to do, with what, and where. But it will not get us to the kind of fine-grained manipulation we need in a kitchen or laundry room: how to do something at the contact/force level, under uncertainty, with slippery or fragile or deformable objects.

For that, we need to collect high-quality robot data—observation-action data: sets of what’s happening and the motions, forces and feedback that lead to a smart robot brain. But there are almost no deployed robots from which to collect that data. 

Here’s the chicken-and-egg problem. To collect that data, we need robots out in the world (in the distribution of homes they’d work in). But there are no deployed robots in homes, and no reason to deploy them until they are much more useful. Paid teleoperation delivers insufficient diversity/quality of data.

What Sunday is doing differently

The Sunday team knows how to do creative, directed research to solve what is fundamentally an integrated systems problem.

At Stanford, Tesla, and DeepMind, the founders contributed to many of the most important breakthrough systems in robotics AI over the past five-plus years: ACT, ALOHA, Diffusion Policy, UMI. They’ve since recruited and enabled an all-star team that can do hardware–AI co-design and deliver full-stack product.

They’ve cracked the central egg.

In just eighteen months, the team has:

• Delivered an end-to-end AI pipeline that lets them go from new task idea to better robot policy with industry-leading iteration speed
• Built a scalable in-home data collection system (Skill Capture Glove) so people can teach robots skills directly in their own, lived-in homes
• Created a unique system to recruit, direct, upskill, and QA a network of in-home data collectors – orders of magnitude more efficient than other approaches
• Developed the ability to train effectively on this demonstration data (Skill Transform)
• Trained ACT-1, a model that is state-of-the-art in dexterity, embodiment, scene and object generalization
• Delivered Memo, their first robot, that will go into the homes of beta partners in the coming months

Building the future, the hard way

Everything about robotics is hard. But as kids, many of us experienced the joy and the expanding sense of self that comes from realizing we can do hard things.

Sunday is doing many hard things at once. Even in prototype form, the output is beautiful.

I love being at the Sunday office. It feels like a team that knows how to do the impossible. Something is always being made or broken, but every day, real progress is made and the future is pulled a little closer. Everyone is working unbelievably hard, but this is a band of happy warriors, because they can already see the path up the hill.

It’s been a blast working with these learning machines, and partnering with Eric Vishria and the incredibly cool band of supporters around Sunday, including Dylan Field , Kevin Weil , Hugo Barra , David Singleton , John Collison and others. Thank you to the Sunday team for letting Conviction be a part of it.

We have so much to do! If you’re a roboticist, researcher, or product builder who wants to work on real robots in real homes – and on extending the frontier, we are hiring the world’s best. Please join us.