Colin Angle's new company Familiar Machines & Magic just announced its first social robot, while hobbyist builders scale autonomous systems independently. Both point to the same shift.
Two announcements arrived within weeks of each other: a stealth Physical AI startup unveiled a social companion robot, and a hobbyist demonstrated room-scale autonomous navigation on a budget Arduino build.
According to IEEE Spectrum, Colin Angle, who co-founded iRobot and led it for decades, stepped down as CEO of iRobot two years before the May 2026 announcement and almost immediately started a new company called Familiar Machines & Magic (FM&M). The company stayed quiet until now. Its first product is described as a physically embodied AI system designed to perceive, adapt, and interact with people in ways that feel natural. Separately, as reported by Interesting Engineering, a YouTuber going by UncleStem built a tortoise-shaped autonomous robot at seven times the scale of a typical tabletop build, still using Arduino-based navigation logic. Both stories landed within two weeks of each other in May 2026.
FM&M's first robot is a pet-like companion for adults. It learns your routines, seeks you out, and is positioned as a Physical AI product rather than a toy.
The framing here is worth unpacking. According to IEEE Spectrum, the FM&M robot is explicitly not a toy, not designed primarily for children, and is intended for adults who want to purchase it for themselves and their families. The product description says it will get to know you, seek you out for attention, and actively help you to pursue an idealized routine. That last phrase is interesting. It implies behavioral nudging, not just passive interaction. The company brought in Morgan Pope from Disney Research, a team known for animatronic character work that feels emotionally resonant. That hiring choice says something about the design priority: perception and natural interaction are being treated as first-class engineering problems.
iRobot succeeded by solving a narrow task, floor vacuuming, extremely well and making it invisible. Angle is now swinging in the opposite direction: a robot with visible social presence, emotional cues, and relationship-building as the core value proposition. That is a fundamentally different design philosophy, and it is a harder problem to get right at consumer price points.
A YouTuber built a room-scale autonomous robot on an Arduino platform. That the navigation still worked at 7x scale is a useful data point about how accessible autonomous mobility has become.
As reported by Interesting Engineering, UncleStem scaled a tortoise-shaped robot to seven times its original tabletop dimensions without losing autonomous navigation capability. The fact that an Arduino-based control system survived that scaling is notable. It suggests the core sensing and navigation logic was not the constraint. What changes at room scale are the mechanical loads, power requirements, and the consequences of navigation errors. A tabletop robot that bumps into something is a minor event. A room-scale robot doing the same hits furniture, pets, or people. The hobbyist community is, in effect, stress-testing the boundary conditions of low-cost autonomous mobility.
Both projects reflect growing confidence that robots can operate in human-inhabited spaces without being human-shaped. That is a meaningful departure from the humanoid-first narrative dominating Physical AI headlines.
Most Physical AI coverage in 2026 centers on bipedal humanoids: Figure, Tesla Optimus, Unitree, and others. But these two projects are doing something different. FM&M is building a pet-like social robot, and UncleStem is building a tortoise. Neither is trying to walk on two legs or manipulate tools with human-like hands. What both share is an assumption that useful or engaging robots can have non-humanoid form factors, operate at human-scale environments, and still deliver meaningful autonomy. The hobbyist project demonstrates accessibility of autonomous navigation. The startup demonstrates that serious talent and investor interest are flowing into social, non-humanoid robots. Those two data points reinforce each other.
Humanoid robots typically need 30 or more degrees of freedom to replicate human motion. A social companion robot with pet-like behavior might need far fewer. Fewer joints means fewer actuators, simpler control loops, lower cost, and more reliability. FM&M has not published specs yet, but the form factor choice likely reflects a deliberate actuator count decision, not just an aesthetic one.
Pricing and behavioral data from FM&M will determine whether social robots can cross the consumer adoption threshold. DIY scaling experiments will keep showing where the real engineering constraints live.
According to IEEE Spectrum, FM&M has not announced pricing or a release timeline for the Familiar robot. That gap is where a lot of the real story lives. Social robots have failed commercially before, including products from well-funded companies with strong technical teams. The difference FM&M is betting on appears to be the combination of Physical AI inference capabilities and the behavioral design expertise from teams like Disney Research. On the hobbyist side, projects like UncleStem's tortoise build serve a different function. They compress the feedback loop between concept and physical test. What the hobbyist community discovers about failure modes at room scale will eventually inform commercial product decisions, whether or not the commercial teams are watching.
Familiar Machines & Magic (FM&M) is a Physical AI startup founded by Colin Angle, the co-founder and former CEO of iRobot. According to IEEE Spectrum, Angle founded the company shortly after stepping down from iRobot two years ago. The company's first product is a social companion robot for adults.
According to IEEE Spectrum, FM&M describes its robot as designed for adults, not children, and positions it as a physically embodied AI system rather than a toy. It is pet-like in behavior rather than humanoid in form, focusing on social interaction, routine support, and relationship-building rather than physical task manipulation.
As reported by Interesting Engineering, YouTuber UncleStem built a tortoise-shaped autonomous robot at seven times the scale of the original tabletop version, and it maintained autonomous navigation. This shows that Arduino-based navigation logic can survive significant physical scaling, though the mechanical and power challenges grow substantially at larger sizes.
Form factor directly drives actuator count, degrees of freedom, cost, and reliability. A pet-like social robot likely requires far fewer joints than a humanoid, which simplifies the control system and reduces the number of actuators that can fail. Fewer mechanical components also lowers manufacturing cost, which is critical for consumer price points.
The track record is mixed. While iRobot succeeded with task-focused robots like the Roomba, socially interactive robots have repeatedly struggled to reach mass adoption despite strong technical teams and significant funding. FM&M's bet appears to rest on improved AI inference capabilities combined with character design expertise from teams like Disney Research.