Colin Angle spent three decades teaching robots to vacuum floors—hardly the stuff of emotional connection. Now, in a move that's either visionary or quixotic (time will tell which), he wants one to make you feel less lonely.
When the iRobot cofounder emerged from stealth on May 4, 2026, at The Wall Street Journal's Future of Everything conference in New York, he brought something decidedly un-Roomba-like: a bear-shaped quadruped robot designed not to clean your home but to keep you company in it. Roughly the size of a bulldog and deliberately abstract—not quite a dog, not quite a teddy bear—the machine represents Angle's answer to a question that has vexed the robotics industry for years: Can a robot be more than a tool?
It's a bold gamble into territory strewn with cautionary tales. Jibo shut down. Anki's Cozmo failed to find staying power. Mayfield Robotics' Kuri never made it past early adopters. The companion robot graveyard is crowded with well-funded, beautifully engineered devices that ended up gathering dust in closets—assuming they made it into homes at all.
Angle thinks he knows why they failed. More importantly, he thinks he knows what to do about it.
The Bear That Isn't Really a Bear
The robot that Familiar Machines & Magic unveiled looks nothing like a vacuum cleaner, which is the point. The company calls it an "abstract bear," a deliberate design choice to sidestep what Angle described to IEEE Spectrum as "mismatched expectations." Make it look too much like a dog and people expect it to fetch. Too human and you're deep in uncanny valley territory.
The machine packs 23 degrees of freedom—engineer-speak for movement that can appear surprisingly lifelike. It features a custom touch-sensitive coat, vision systems, microphone arrays, audio capability. What it doesn't have: a screen. It won't talk to you in the conventional sense. Instead, it makes animal-like sounds, understands what you say, and communicates primarily through whole-body movement and gesture. Think less Alexa, more golden retriever.
Two prototype units shown at the conference—dubbed Daphne and Winston, because even robots need names—demonstrated behaviors the company spent two years engineering: reading a room's mood, responding to touch, following owners around the home. The robot is designed to accompany you on walks outside, not just patrol your kitchen. Whether it can actually pull that off in Boston winter or a crowded Manhattan sidewalk remains an open question.
When Hollywood Writes the Code
Here's where Familiar Machines diverges from your typical robotics startup, and it's either brilliant or deeply strange: they hired screenwriters.
The company worked with Hollywood talent to craft character arcs and day-in-the-life stories for their robots—actual narrative scripts, the kind you'd write for a Pixar character. Those narratives then get expanded through generative AI into training material for the on-device behavior model. It's an approach that blends entertainment industry character development with machine learning. Less about engineering personalities from scratch, more about scripting them first and letting the AI fill in the gaps.
The Disney influence runs deeper than storytelling technique. Creative Director Morgan Pope joined from Disney Research, where he worked on bipedal and acrobatic robotics—the kind of things you might see at a theme park, engineered to withstand millions of interactions. Company materials emphasize team members from Disney Imagineering alongside veterans from iRobot, Boston Dynamics, MIT, Amazon, Bose, and Sonos. The company maintains offices in Boston, Los Angeles, and Hong Kong. That LA presence isn't accidental; it signals deliberate access to the entertainment industry's creature-effects ecosystem.
Tombot's Jennie robotic puppy took a similar path, partnering with Jim Henson's Creature Shop for design and animatronics expertise when it debuted at CES earlier this year. The convergence of robotics and Hollywood talent isn't coincidence. It reflects a growing recognition that emotional connection requires more than sensors and servos—it requires an understanding of character, presence, and what makes something feel alive rather than merely animated.
The Privacy Argument (And Why It Matters Now)
Familiar Machines runs its entire AI stack on the robot itself. The custom small multimodal model combines vision, audio, language, and memory for real-time social behaviors without continuous cloud streaming. Optional connectivity enables updates and features, but the core intelligence lives locally—on the device, in your home, theoretically beyond the reach of corporate servers.
Angle framed this as "Roomba Rules" for data governance in an interview with Business Insider. After building vacuum cleaners that mapped millions of homes, he's acutely aware of privacy concerns around always-on devices. The iRobot/Amazon acquisition saga—which ultimately fell apart over regulatory scrutiny—only sharpened that awareness.
The on-device approach also enables faster response times, which matters more than you might think. When trying to create the illusion of presence and awareness, latency kills the spell. A half-second delay between your gesture and the robot's response transforms a companion into an appliance.
The technical architecture includes what Angle described to IEEE Spectrum as a reinforcement learning-based unified motion model optimized for expressivity rather than load-bearing tasks. The quadruped locomotion isn't about carrying groceries; it's about appearing attentive, curious, responsive. This thing won't fetch your slippers, but it might tilt its head when you're having a bad day.
Crowded Field, High Stakes
Familiar Machines isn't entering a vacuum, so to speak. CES this year saw multiple companion robot debuts, including Ecovacs' LilMilo. Sony continues selling Aibo at a premium—$2,899 when it launched in the U.S. in 2018, still not what anyone would call cheap. KEYi Tech's Loona targets the budget segment around $399. Japan's LOVOT maintains an active, if niche, presence in its home market.
The difference, perhaps more than Angle would care to admit, is pedigree. He cofounded iRobot in 1990 and spent decades scaling manufacturing and navigating consumer expectations. Cofounder Chris Jones served as iRobot's CTO. Ira Renfrew, Chief People and Product Officer, brings experience from Amazon and autonomous vehicle startup Outrider. The advisory board reads like a who's who of robotics: Marc Raibert of Boston Dynamics, MIT's Cynthia Breazeal (who founded Jibo and knows firsthand how companion robots can fail), and USC's Maja Matarić.
That roster reflects both deep robotics expertise and intimate familiarity with social robot failure modes. They've seen what doesn't work. Whether they've figured out what does is another matter.
The Economics of Robotic Companionship
Familiar Machines plans to charge an upfront purchase price plus a monthly subscription "comparable to owning a pet," Angle told The Boston Globe. Exact pricing remains undisclosed, which is either strategic positioning or they haven't figured it out yet.
For context: the average American spends somewhere between $1,000 and $2,000 annually on dog ownership, according to various pet industry estimates. Whether consumers will stomach that for a robot remains deeply uncertain. The subscription model signals ongoing content, behavior updates, and cloud features—an attempt to sustain engagement beyond the initial honeymoon period where previous social robots stumbled and never recovered.
An SEC filing from December 2024 confirmed the company has raised $15 million. The team of nearly 40 employees—split between Woburn, Massachusetts and California—is working toward a 2027 commercial launch. A waitlist is live on the company's site, though specific ship dates and initial markets haven't been announced. The company is targeting manufacturing through a contract factory in Asia, leveraging the production scaling expertise Angle developed at iRobot.
What We Don't Know Yet
What Familiar Machines showed in May was explicitly "not a commercial product launch," according to its own press materials. Demos are controlled environments. Real-world durability, battery life, and long-term engagement remain to be proven—especially that last one, which is where companion robots have consistently failed.
The company positions this as creating a $5 trillion "Consumer Physical AI" market, which is the kind of number you put in pitch decks to excite venture capitalists. It's vision, not validation. Grand projections about massive markets have preceded many a robotics disappointment.
Angle's belief that connection-oriented physical AI represents a bigger near-term opportunity than humanoids sets him apart from much of the industry, where bipedal robots attract most of the attention and funding. He's skeptical of bidirectional human-style speech in the near term, wary of uncanny valley problems, and betting that expressiveness through motion can carry more emotional weight than conversation. It's a theory. A reasonable one, but still a theory.
Whether households actually want a bear-sized robot companion—whether they'll pay for one, keep it charged, integrate it into daily routines rather than shove it in a closet after three weeks—will ultimately determine if this product category exists or if Familiar Machines becomes another well-intentioned entry in the social robotics graveyard.
Angle has earned credibility through decades of commercial robotics success. The Roomba became a household name and a genuinely useful product. He'll need every bit of that credibility, and probably some luck besides, to convince consumers that this time the companion robot will stick around.
The difference between a beloved household presence and an expensive paperweight might come down to something harder to engineer than locomotion or AI: whether people actually miss it when it's not there.
