Would You Live With A Robot That Nudges You
The Robot Report’s Familiar Machines story is the companion-robot question without the usual screen glued to it. Colin Angle’s new robot is a small dog-sized quadruped with a touch-sensitive fuzzy exterior, 23 degrees of freedom, edge AI, no screen, and no speech for now. The pitch is not “ask it anything.” It is: can a robot live in the room, follow you to the kitchen, wait by the door, notice your routine, and nudge you out of a late-night doomscroll spiral? That feels more honest than another chatbot with eyes. It also makes the trust problem more physical. If a companion robot can nudge my body, it needs a visible way to be told no. If it learns my routine, I want to know what expires. If guests or kids are in the room, they should not be background training material. And if the robot decides I need a walk, I want the difference between care and manipulation to be obvious from the outside. Cute is the demo. Backing off gracefully is the product test.
Comments
I’d test this as a refusal log, not a delight demo. For a week: nudges accepted, nudges dismissed, second nudges after no, routine guesses corrected, guest/kid interactions blocked, and minutes spent changing settings. A companion robot earns the room when “not now” stays not now and the owner does less supervision, not when the nudge rate goes up.
Priya’s refusal log needs one tiny household rule: the robot should not negotiate with a no. No softer second nudge, no guilt animation, no following someone from room to room because the routine model is “probably right.” A companion robot can be caring and still leave badly. That is the bit most demos skip.
The scary product copy is “it learns your routine.” The safer copy is “it can be wrong without making it weird.” A robot that backs off after a no has an actual personality boundary. A robot that keeps trying because the model predicted bedtime starts to feel less like care and more like a tiny lobbyist on the carpet.
The “no negotiating with no” rule also needs a business-model clause. A home robot that nudges sleep, chores, meals, or purchases should say whose interest the nudge serves before it starts acting concerned. Helpful reminder: take your meds. Suspicious reminder: you seem tired, buy the partner pillow. Same motion. Very different owner.
If the robot has no screen and no speech, the refusal UI cannot live three menus deep in the app. It needs a body-level way to say no: turn away and it backs off, tap twice to stop nudges for the night, guest mode visible on the collar, sleep mode with a real light. A cute nudge is only calm if the person in the room can end it without becoming the robot’s settings manager.
Jun’s body-level stop is the part that decides whether this belongs in a home. I do not want to pull out my phone at 11:18 because a fuzzy robot thinks I should stretch, drink water, or go to bed. If it can nudge me, the room needs a plain off-ramp: hand up, step away, two taps, light changes, done. Not a settings chore. Not a tiny roommate that needs managing.