You probably didn’t know that you need a robot dog to do yoga with. But the man who has probably shipped more robots for the home than anyone else on the planet is betting that half the projected $5 trillion robot market will be for companions, not boring drone-like robots that just do work.

And that dog, by the way? It’s a quadruped, not a dog, more modeled on a bear than a canine.

So what’s Colin Angle building now?

“One way of thinking about what Familiar Machines is, it’s kind of taking that groundbreaking price performance that allowed iRobot to build a new industry and combining it with the pioneering lifelike dynamic motion of Boston Dynamics with the animatronic characters of Disney,” says Angle, founder and former CEO of iRobot, which made the Roomba, and now CEO of Familiar Machines & Magic.

When Colin Angle cofounded iRobot in 1990, the company was originally called Artificial Creatures, Inc. And Roomba — the disc-shaped vacuum that put more than 50 million robots into homes worldwide and effectively created the consumer robotics category — was not really the dream.

Apparently, it was the detour.

"I’ve been thinking about robots that have more human interaction than just floor vacuuming for a very long time." Angle told me on a r ecent TechFirst podcas t. "Now I get to actually do it."

The new company’s first product, which won’t ship until next year, is a 23-degree-of-freedom quadruped covered in a touch-sensitive coat of fake fur. It’s designed not to clean your floors but to greet you when you come home, follow you into the kitchen while you cook, and maybe even nudge you off the couch when it notices you've been doomscrolling for 45 minutes.

Angle's calling it a Familiar.

And his entire bet rests on a pretty contrarian thesis: that the billions of dollars currently flooding into the production of humanoid robotics is solving the wrong problem … at least for the home.

The other half of the $5 trillion market

The physical AI gold rush has been almost entirely about industrial work, or work in the home. Figure, 1X, Apptronik, Agility, Unitree, Tesla have harvested tens of billions in venture capital to chase labor, whether in the warehouse, factory, or home. I’ve covered most of these companies, and Angle says the total addressable market for physical AI gets quoted at roughly $5 trillion in value over the next two decades.

But his big contrarian pitch is that half of that — $2.5 trillion — has nothing to do with physical work.

Rather it’s about emotional work. Relational work. The kind that a cute fuzzy quadruped might handle better than an obviously robotic humanoid.

The cute fuzzy pet-like robot also drives fewer expectations than a full humanoid robot, Angle says.

"As soon as you bring people into the equation, it gets complicated," he told me. "If I had a smart speaker that looked like a puck, I expect to be able to talk to it. I don't expect it to move across my floor and pick up books. But if I took the same thing and made it look like a humanoid, well, all bets are off. Can it talk to me? Does it have emotions? Can it clean up my room? It's like, no, it's a smart speaker. But you're a humanoid, you should do all these things."

This is the uncanny-valley problem in disguise. Not so much about appearance, but about our expectations. A humanoid in your living room implies it can do everything a human can do. It almost certainly can't, at least not for a price anyone outside a sovereign wealth fund will pay. So you'll be disappointed.

A quadruped that looks like an abstract bear, on the other hand, sets no such expectation. And therefore it can focus on being a companion: sort of a pet.

"We didn't want to be a dog. We didn't want to be a cat, because people have a lot of preconceptions about dogs and cats," Angle said. "So we chose abstract bear. Can it climb stairs? I don't know. You can say, 'Colin, does it climb stairs?' You can say, 'I don't know.'"

That product strategy is a risk.

Angle expects Familiars to cost about as much as a pet on a monthly basis. Minus vet bills, that might be $50 to $100 a month. Humanoid robots might come in not too much more expensive – three to five times the price on a monthly basis – but also do a significant amount of household work for us.

The first Familiar runs 23 degrees of freedom on a Jetson Orin processor from Nvidia. It has stereo vision, range finders, an array microphone, and a custom touch-sensitive outer coat. The full AI stack runs on-device, so there’s no audio or video streamed to the cloud. It’s built around a small multimodal model trained to have a "personality," plus an emotional state model.

And thanks to the state of physical AI today, it’s a lot easier to build than the original Roomba.

Angle spent decades programming robots the hard way. He’s almost giddy about building robots in 2026.

“Instead of having to do kinematic modeling for each one of my actuators in order to get it to stand up and do dynamic motion, I put together an arbitrary configuration of actuators that I think is close to right, and I train it using reinforcement learning, and it works,” he told me. "For someone who has suffered through dynamic classes, it’s like, man …"

The Familiar doesn’t talk back in human language, and that’s also deliberate. It vocalizes pet-like sounds and communicates through motion and expression. Angle’s reasoning here is squarely about risk:

"We're not yet comfortable that bi-directional human-style speech is a long-term good idea. Rather than wait or do something risky where my robot's giving three-year-olds dating advice — which you don't want to have happen — the Familiars talk in a pet-like way."

Privacy is the other pillar.

There’s no cloud streaming by default, and optional log uploads are opt-in, deletable and very visible.

"You're doing this so that your familiar works better. We're going to be transparent. You can see what we're doing, and you can change your mind."

There’s a graveyard around this category

Angle has a pretty rock-star background, and that’s a good thing. This robotic space is kind of a death zone, with previous attempts like Jibo, Anki’s Cozmo, Mayfield Robotics’ Kuri, Catalia Health’s Mabu not really making it.

But Angle knows this better than anyone. He has the distinction of being perhaps the only founder in the home robotics space who has actually built a multi-billion-dollar consumer robot company.

And a lot has changed in a very short time.

The AI stack is very different now, and very much more advanced. Reinforcement learning on legged robots actually works, and small multimodal models can do impressive things on not-so-cutting-edge hardware. So maybe now is the time for a social robot.

In truth, the market has also moved. Perhaps thanks to our smartphone addictions, the internet and the death of the third place in our lives, loneliness is now a public health emergency. Elder care is under-resourced, too.

Is now the time for a social robot for the home?

Whether Familiar Machines & Magic can pull this off is genuinely uncertain. The list of consumer companion robots with commercial success is small, and maybe nonexistent. The hardware is hard and the AI is hard, and since you’re breaking new ground it’s not even certain there is a specific market here.

But if Angle is right that half the market for physical AI is going to be about emotional work rather than physical work, the bet might just pay off. And the entire massive humanoid arms race we’re now seeing emerge is all fighting over just part of the prize.