Angle, Colin
▶ Watch on YouTubeCore ideas
Angle argues Asimov's Three Laws of Robotics require a level of self-awareness and understanding of consequences that makes them irrelevant as an actual engineering bar for decades to come — Roomba arguably satisfies them in spirit (designed to help, not harm, and to be durable) not through any explicit ethical reasoning but simply because being safe and helpful aligns with being a good consumer product. He frames robots' broader societal role in demographic terms: as the population ages, there won't be enough younger people to care for the elderly, and robots will increasingly be the difference between "gut-wrenching declines" in independence and a future where people retain control over their time and living standards, with the home specifically being iRobot's chosen arena for incremental progress toward a home that actively maintains and cares for its occupants.
Angle traces iRobot's foundational insight — learned only after early ventures in high-tech, lab-grade robotics failed to find a market — that groundbreaking technology alone doesn't constitute a viable business; a robot must deliver value to the end user clearly and sustainably in excess of its cost, not marginally. He credits early market research done with Johnson Wax, which sent researchers into people's homes simply to observe what tasks people did reluctantly but too frequently to ignore, as key to identifying floor cleaning as that first "obvious value" wedge — a discipline he argues explains why other well-funded, technically brilliant robotics companies (Anki, Jibo, Mayfield Robotics/Kuri, SciFi Works, Rethink Robotics) ultimately failed: they built compelling robots without first nailing down a clear, sustainable value-versus-cost equation, a challenge he compares to the toy industry's brutal ~85% failure rate for products even with real merit, especially in social-companion categories where the "magical user experience" bar remains unclear.
He describes two enabling manufacturing/technology shifts that made affordable home robots possible: first, moving away from machined aluminum parts (priced by labor-hours) toward injection-molded plastic at scale, where unit cost is driven down to roughly the cost of the raw plastic material itself, unlocked by better 3D CAD tooling; and second, choosing vision-based sensing over tactile "skin" sensors or lidar, betting early on the trajectory of Moore's Law and the smartphone industry to eventually make machine-vision-capable processors affordable at consumer price points — a bet Angle says paid off roughly two years before the interview. He draws a direct comparison to the lidar-versus-camera debate in autonomous vehicles (referencing Elon Musk's "lidar is a crutch" position), noting iRobot's environment is actually harder visually than roads (infinitely varied floor textures, colors, thresholds, and obstacles) but far more forgiving in terms of safety consequences (a slow, lightweight robot bumping a foot is comic, not dangerous) — concluding that vision will likely be the primary sensing modality for both domains, though he's open to cheap lidar as a supplementary backup sensor in vehicles given the much higher safety stakes there.
On the current and future trajectory of home robotics, Angle frames Roomba's autonomous localization and mapping capability as only the foundational first step, with the next major direction being counterintuitively "less autonomy" in the sense of tighter human-robot collaboration: rather than a fully independent automaton, the goal is a robot that listens to context-specific natural-language instructions (e.g., "clean up the kitchen" or reacting to "I just dropped flour by the fridge"), maps that utterance against its existing semantic understanding of the home's layout, and acts accordingly — a partnership model where the robot has a sensible default routine but remains continuously responsive to specific human direction. He connects this evolving spatial and semantic home-mapping capability directly to the long-standing robotics dream of manipulator arms (citing Rosie, Robby the Robot, and Data as science-fiction touchstones): an arm is useless without knowledge of where things are, so the current wave of home-understanding technology is what will eventually make useful physical manipulation (like fetching a beer from a known refrigerator location) actually tractable.
On privacy, Angle describes iRobot's approach as central to the company's long-term viability given its ambition to be trusted with an increasingly detailed model of customers' homes: a public "privacy manifesto" committing never to sell customer data, global (not just jurisdiction-mandated) adherence to GDPR-style standards, on-device processing of raw visual/sensor data (so images never leave the robot), and user control over whether semantic home-map data is even generated or retained, with the ability to have it forgotten on request. He argues the deeper problem isn't winning individual customer trust so much as the absence of any commonly understood, verifiable privacy standard in the market — noting candidly that prominently advertising iRobot's privacy commitments on packaging would likely reduce sales rather than increase them, since consumers currently express concern about breaches in the abstract but don't factor privacy practices into actual purchasing decisions — and calls for an industry-wide, comprehensible "grade" or standard that could let good-faith privacy practices actually become a competitive advantage rather than a costly, unrewarded commitment.
Asked about robots achieving human-level intelligence, Angle is direct that this won't happen in his lifetime, but argues a robot can still provide a genuinely "meaningful conversation" without anything resembling human-level cognition — provided the value proposition is correctly framed around what the human learns or gains (information about their pet, home energy efficiency, planning advice) rather than an illusion of the robot's own feelings or wellbeing, which he contrasts with a trivial "make the robot happy" interaction that could be replicated by a button that makes it smile. Closing on a reflection about emotion in robotics (prompted by his choice of Data from Star Trek as a hypothetical robot companion), Angle proposes that as robots become more capable, they will likely become more emotional rather than less, since he believes emotion serves a genuinely useful computational role in social, competitive, or dangerous situations involving incomplete information — situations he argues are far too common, and too poorly suited to pure logical resolution, for a purely rational agent to navigate well.
Excerpts
On why technical brilliance alone doesn't make a viable robotics business:
“Technology alone doesn't equal a successful business — we need to go find the compelling need where the robot that we're creating can deliver clearly more value to the end user than it costs... not a marginal thing where... maybe we can hold our breath and make it work — clearly more value than the cost.”
On why vision beat tactile sensing and lidar for home robots:
“Instead of trying to build skin... why don't we focus on vision, and how many of the problems that face a robot trying to do real work could be solved with a cheap camera and a big-ass computer?... between injection-molded plastic and a camera with a computer capable of running machine learning... I could build an incredibly affordable, incredibly capable robot.”
On why privacy commitments alone aren't yet rewarded by the market:
“If I put on the package of my Roomba the privacy commitments that we have, I would sell less than if I did nothing at all, and that needs to change... it's not a question about earning trust — we need to figure out how to have a comfortable, understandable standard applied to privacy that customers can trust and use in their buying decisions.”