A San Francisco startup emerged from stealth yesterday with a home robot that costs $3,555, learns a new chore in about 30 minutes, and can lift a grand total of 3.3 pounds. It’s not particularly pretty – although it has a cute hat and fresh colors – and its grippers are devastatingly basic. Flourish 1’s base looks like a chopped-off garbage can with wheels (or R2-D2?) on top of which the company built a body of sorts with a head and two arms, and that body moves up and down on a pillar attached to the base.

All of this is exactly on purpose, including the 3.3 pound weight limit.

This is a home robot designed to be practical and trainable, but above all, accessible for fairly average people.

“We’re building a robot specifically for the home. Most humanoid companies build general-purpose robots for factories, outdoors and homes, which means high payloads and industrial-grade hardware,” Flourish founder and CEO Antoine Marcel told me in an email interview. “We don’t need that. Flourish 1 carries 3.3 lb, isn’t waterproof, and is designed around what people actually need at home. That lets us simplify the hardware dramatically and reduce cost.”

The company is opening pre-orders for the first 50 units of Flourish 1, with deliveries starting in December 2026. Owners will be able to teach it household tasks in a mobile app, no coding required, simply by physically demonstrating the task with their phone. Once it’s learned a task, Flourish 1 can repeat it and run it on a schedule: picking up shoes, clothes and toys, wiping tables, watering plants, fetching things and even clearing a cat litter box.

It’s not fast, and it’s a bit herky-jerky. Plus, only a recalcitrant teen co-opted unwillingly into home cleaning would call moving a cloth over a table, as you can see in the launch video above, “cleaning.”

But it’s incredibly cheap.

For context, 1X’s NEO, probably the best-known humanoid robot purpose-built for home use, costs $20,000 or $499 a month. Figure AI is building Figure 04, and increasingly signaling that home is an important market for the company. Flourish 1 is much cheaper than either of them, and joins a very recent avalanche of very cheap robots from small teams for the home and enterprise lately. Nori Robotics’ A3 sells for just $1,688 , not-so-small Unitree’s R1 starts at $4,900 , and last week Feather Robotics launched a $29,990 wheeled robot aimed at work, not play.

Parts are increasingly cheaper and far easier to get: actuators, LiDAR, cameras and compute that were specialty items a few years ago are now close to commodity, and Goldman Sachs found that humanoid manufacturing costs fell 40% in a single year thanks to scale and commoditization. The brains are getting better, too, and a lot of them are free: open models like Physical Intelligence’s π0 and Nvidia’s GR00T mean a small startup no longer has to build its physical AI from scratch, and can instead fine-tune something state-of-the-art, which is exactly what Flourish is doing.

And there’s simply more talent. Engineers who cut their teeth at Tesla, 1X, Figure and a dozen other well-funded robotics companies are now spinning out to start their own. Put it all together and you get something that would have been impossible even two years ago: a handful of people building a useful home robot for less than the price of a used Honda Civic.

Flourish 1 stands 3’7” (110 cm) tall, weighs 44 pounds (20 kg), and runs for up to 12 hours on a charge, the company says. It can move up on its base to reach higher off the floor.

Three pounds is plenty for socks, shoes, toys and most of the flotsam and jetsam that accumulates on the floor of a family home. It’s not enough for a full laundry basket, a stack of dinner plates or a bag full of groceries. (A liter of water weighs 2.2 pounds, so plant watering will happen with a modest watering can.)

Marcel is candid about where the robot is today.

“Today it can put shoes away, pick clothes up and put them in a laundry basket, throw away trash, wipe a table or water plants,” he says. “It works reliably, but still relatively slowly.”

The press release also lists loading the dishwasher among the chores owners can teach. Marcel says unloading it, along with setting the table and food preparation, is what’s next.

The long-term goal is bigger.

“Ultimately, we want the robot to understand where everything belongs and regularly do a full reset of your home,” Marcel says.

That’s the dream of every parent who has ever stepped on a Lego at 11 PM, of course.

How Flourish plans to get there is where it diverges most from the big players. Figure, for instance, is betting on massive pretraining to build a general physical AI model that works anywhere. It recently showed its humanoid jumping from 9% to 56% zero-shot success on chores in homes it had never seen before: very impressive, and still just the beginning.

Flourish is going the other way.

“We don’t try to build one universal AI model for every home. We start from state-of-the-art pretrained models and fine-tune them on your home, your objects and your way of doing things,” Marcel told me. “We also split tasks into workflows. For picking up a sock, for example, navigation and object detection get the robot to the sock, physical AI handles the difficult grasp, and navigation takes it to the laundry basket.”

In other words, the expensive, finicky part of robot AI, the vision-language-action model that actually controls hands in the physical world, only gets used for the part of the job that truly needs it. Getting across the living room is a solved problem: robot vacuums have been doing it for two decades. Grabbing a crumpled sock – as easy as it is for us meatbag humanoids – is not.

“That means we use physical AI only where it’s really needed, which makes training much faster and more reliable,” Marcel says.

It’s brutally pragmatic, just like Flourish’s ethos on hardware, and it’s probably how a small startup ships a useful robot in 2026 rather than 2029. The tradeoff is generalization: a robot fine-tuned on your house might not do as well when you rearrange the furniture.

Marcel acknowledges that.

“It doesn’t just memorize a trajectory. The pretrained models and our dataset augmentation help it handle changes in object position, lighting, clutter and environment,” he says. “If a VLA policy starts failing, higher-level vision/language models can reassess the scene and help the robot retry or reposition. There are still limits with large changes in the environment, and improving generalization is a major focus for us.”

On safety around kids and pets, Marcel says the robot “continuously monitors its surroundings with LiDAR and cameras, and the arms automatically stop if a person enters their reachable area.” The arms are also deliberately lower-force than those on industrial humanoids, he adds.

And unlike 1X, which has said NEO buyers can bring in remote human operators for tasks the robot can’t yet handle, Flourish says it isn’t relying on teleoperation.

“The robot is designed to operate autonomously once a task has been taught,” Marcel says. “We can teleoperate it for fun, but the robot is designed to work fully autonomously.”

Flourish’s pre-seed round was led by Families Fund , founded by Kartik Sathappan. The company didn’t disclose the size of the round. Marcel is a repeat founder with a prior exit, and previously raised money from a fund backed by French billionaire Xavier Niel.

“Too often, technology is used to distract kids so parents can do chores. Flourish does the chores so parents can play with their kids,” Sathappan said in a statement.

Marcel’s pitch is similar.

“The point isn’t a robot that picks up your shoes,” he said in the statement. “It’s what you get to do because you didn’t have to.”

Right now, Flourish 1 is small, slow and limited — but it costs $3,555, ships in December, and doesn’t need a stranger in a VR headset to pick up your socks. Fifty early buyers are about to find out if that’s enough.

But I’m more excited about Flourish 2, or 3, or 7. This trend of small companies building humanoid robots that would have been impossible just a few years ago at prices almost anyone can afford has real legs. Or wheels.