Can a Home Robot Really Fold Laundry Better Than You?
Memo is an autonomous home robot that folds laundry with 99.1% success in homes it has never seen, learning household chores from millions of real human routines.
Photo source:
Act-2
A Garage, Two Researchers, and a Pile of Laundry
Sunday Robotics began the way Silicon Valley
stories are supposed to begin. Two Stanford-trained roboticists, Tony Zhao and
Cheng Chi, met after reading each other's research papers, then started
building a robot in a garage with 3D printers running around the clock. Their
company, based in Mountain View, California, grew from two people in an
apartment to a team valued at over a billion dollars in roughly two years. The
machine they built is called Memo, and its defining achievement so far is one
of the most ordinary tasks imaginable: folding laundry.
That choice of task is deliberate. Laundry is
deceptively brutal for machines. Garments crumple, tangle, hide their own
edges, and never appear the same way twice, which is why home robots have
historically dazzled in staged demos and failed in real houses. Memo's newly
published results attack exactly that gap. Across 785 fully autonomous attempts
in homes the robot had never seen, it folded and stacked garments successfully
99.1 percent of the time, with no setup, no adjustments, and no human help.
Taught by Ten Million Human Moments
The unusual part of Memo's story is how it
learned. Instead of training a robot in a laboratory, Sunday Robotics built a
wearable device called the Skill Capture Glove and paid people to simply do
their chores while wearing it. The gloves, which cost about 400 dollars a pair,
recorded how real people move, clean, fold, and organize in more than 500
actual homes, producing roughly 10 million episodes of genuine household
routines. That library of human behavior became the robot's education.
The results go beyond memorization. In the
company's latest research, the underlying model, called ACT-2, learned a
completely new folding technique from a single demonstration and then applied
it to garments it had never encountered. The scope of the laundry evaluation
was equally broad, covering nine garment types from T-shirts and polos to
pants, leggings, and blouses, in sizes from XXS to 8XL, starting crumpled in
piles on beds, in baskets, or on the floor. The average fold took just over two
minutes, and independent annotators graded the quality at 4.72 out of 5, with
the vast majority of folds rated four stars or higher.
Built Soft, Built Stable, Built for Households
Memo's body reflects its intended workplace.
Instead of walking on legs like the humanoid robots dominating headlines, it
moves on a rolling base, a choice made for balance, lower weight, and safety,
since the robot remains stable even if power is lost. Its exterior is clad in
soft silicone, designed to feel approachable in kitchens and living rooms
shared with children and pets. The design proves its worth in the evaluation
footage: Memo keeps working steadily when a child interferes with the laundry mid-fold,
when someone deliberately disturbs the garment, and in rooms both dark and
bright.
The robot also uses its whole body as a tool.
It repositions itself around the bed, adjusts its height, and leans into its
workspace, which is how the same machine handles a baby shirt, an oversized 8XL
top, and a large towel. When a garment falls to the floor, Memo treats the mess
as a situation to recover from rather than a failure, retrieving the item and
continuing the fold. Laundry is only the first skill through this standard; the
same underlying model is already practicing vacuuming, organizing toys,
fastening zippers, turning pants right side out, clearing tables, loading
dishwashers, and preparing coffee.
What Exists Today and What Comes Next
Honesty about status matters here. The memo is not
for sale, and no public price exists yet. This fall, Sunday Robotics begins its
Founding Family Beta, placing individually numbered robots into 50 selected
households whose feedback will shape the product's development. The company
frames its progress through an unusually transparent standard it calls a Solve,
publishing not just success rates but the exact scope of conditions tested and
every evaluation video, an approach meant to separate real capability from
staged demonstration.
The broader significance is the milestone
itself. Robotics has produced impressive one-off demos for decades, while
reliable performance across the endless variation of real homes remained out of
reach. A machine folding strangers' laundry, from crumpled piles, in unfamiliar
rooms, at a 99.1 percent success rate, is evidence that the gap between demo
and daily life is closing. Whether Memo becomes a common household appliance
will depend on price, durability, and years of real-world performance still ahead.
What its laundry results already show is that the era of genuinely useful home
robots has moved from prediction to schedule.
You have exceeded your free limits for viewing our premium content
Please subscribe to have unlimited access to our innovations.
Get The Latest In Innovation
Join Our Newsletter
Where bold ideas find bold readers.
