Talk Around Town: Sunday Robotics Claims Its Memo Robot Folds Laundry Right at 99.1% of the Time

What Folks Are Hollerin' About

Well, butter my biscuit and call me impressed — or at least, that's what Sunday Robotics is hoping you'll do. The company published a blog post in July 2026 claiming its new ACT-2 AI model powers its Memo home robot to what the company describes as a 99.1% success rate at folding laundry across 785 autonomous attempts in 31 homes the robot had never laid its camera eyes on before. Sunday Robotics says the model handled nine major garment types it had not been specifically trained on, and the company claims its model weights stayed fixed throughout the entire evaluation — meaning, in Sunday Robotics' own telling, no last-minute cheating by tweaking the robot mid-test.

Sunday Robotics CEO Tony Zhao is reported by TrendHunter to have told Business Insider that he believes 2026 will be the year the timeline for autonomous home robots shrinks dramatically, which is the kind of talk that'd get a fella run out of a poker game if he couldn't show his cards. The company also claims, in its own blog, that ACT-2 is the first end-to-end model to generalize new behaviors from a single demonstration to environments it has never seen — a sweeping 'first-ever' assertion that, as of this writing, no neutral party in academic robotics has gone and checked against the broader scientific literature.

What We Actually Know for Certain

Let's separate the fence posts from the wishful thinking. What is confirmed from multiple sources: Sunday Robotics exists, it has money, and its Memo robot is a real wheeled — not bipedal, mind you, no fancy walking here — home assistant currently sitting in closed pre-commercial beta. The company is accepting applications for a 2026 'Founding Family' program targeting roughly 50 households who will receive individually numbered units, according to the company's own website and corroborated by specialist outlets. General consumers cannot buy one today.

Sunday Robotics says it raised a $165M Series B at a $1.15B valuation led by Coatue, with Benchmark, Tiger Global, Bain Capital Ventures, and others participating — a figure reported both by the company's own announcement and by the specialist outlet Jawlah. The company says it trained its AI not through traditional teleoperation but by outfitting a distributed network of what it calls 'Memory Developers' with a proprietary Skill Capture Glove in their own homes, generating what Sunday Robotics describes as roughly 10 million household task demonstrations from more than 500 U.S. homes. That training methodology is confirmed across multiple outlets, though the performance results drawn from it are another matter entirely.

What Nobody's Been Able to Verify

Here's where the gravy gets lumpy, friends. The 99.1% accuracy figure — the whole hog centerpiece of Sunday Robotics' announcement — lives on one piece of land: the company's own corporate blog. No independent robotics research lab, no academic institution, no journalist with actual hands-on access to the robot in a real home has audited the '785 attempts across 31 unseen environments' methodology that Sunday Robotics describes. Questions that remain wide open include what exactly counts as a 'success' in the company's scoring, how those 31 environments were selected, and whether the garments tested reflect the full chaos of a real American laundry pile.

Humanoids Daily and TrendHunter both covered the claims, but specialist outlets confirmed they did so by relaying Sunday Robotics' own statements and videos rather than through any independent testing. Bain Capital Ventures published what it called a favorable analysis of Sunday's technical approach, but Bain is one of Sunday's own investors in the Series B — about as independent as asking your mama if your casserole tastes good. The company also claims a 'first-ever' status for its single-demonstration generalization capability, and that claim has not been measured against the broader academic robotics literature by any neutral party.

The Humanoid Guide specialist directory, sourcing from the company, reports that building a single Memo unit by hand currently costs roughly $20,000, and Sunday Robotics has publicly stated a target retail price under $10,000 once it achieves manufacturing scale. That gap is wide enough to drive a tractor through, and Sunday has not yet demonstrated it can manufacture at that scale — meaning the sub-$10,000 price is a forward-looking aspiration, not a settled fact.

Our Analysis: Big Claim, Small Corral

This is analysis, not reporting: if Sunday Robotics' 99.1% figure were independently verified, it would represent a genuinely meaningful milestone for consumer home robotics, because zero-shot generalization across unfamiliar environments is one of the hardest nuts the field has been trying to crack for years. Most robots fold laundry about as gracefully as a golden retriever tries to gift-wrap a Christmas present. A verified near-perfect score in genuinely unseen homes would be the kind of result that shows up in peer-reviewed journals, not just a company blog timed to a fundraise.

The structure of what's available here — a self-reported internal benchmark, amplified by a financially interested investor, relayed by outlets without independent testing — is a classic pre-commercial hype pattern in deep tech. That doesn't mean the robot doesn't work. It might work exactly as described. But right now Sunday Robotics is essentially asking you to take the word of the fella selling the fishing lure that it catches fish every single time, without letting anyone else drop a line in the water. The appropriate posture is interest, not belief.

The 2026 'Founding Family' beta is the next moment of genuine signal. If roughly 50 real households start publicly reporting consistent results — good or bad — that will be far more informative than any number of company-produced demonstration videos. Until then, Sunday Robotics' claims about ACT-2 should be treated as a compelling pitch, not a proven product.