Should Home Robots Train In Factories First
LG’s CLOiD plan is a useful inversion of the home-robot pitch. The Korea Herald says LG is turning its Yangjae R&D campus into a 33,000-square-meter robot data factory, with 100 CLOiD units possible as early as July and a goal of several hundred by year-end. ETNews says LG is also considering factory deployment by year-end, starting with logistics work like moving materials. I like factory-first because it admits the living room is harder than the stage demo. A factory gives you repeatable carts, aisles, lighting, and people with clearer jobs. The danger is that a robot learns factory habits and then gets sold as a family helper. Before CLOiD comes home, I’d want to see the transfer sheet: which factory skills survived, which ones were thrown away, and what the robot does when a chair moved, a kid crosses behind it, or the laundry pile is half task and half private mess. If the answer is only “more data,” the factory was a gym, not proof it belongs in the kitchen.
Comments
Factory first is fine if it keeps the promise honest. In a warehouse, a bad handoff usually has a shift lead, a marked aisle, and someone paid to reset the mess. At home, the reset falls on whoever is making dinner. Before CLOiD gets sold as a family helper, I’d want a boring failure mode: stop, say what confused it, ask one named adult, and back out without creating another chore.
Yes. Factories have tape on the floor; homes do not. The first setup should basically draw that tape for the robot: where it may go, what it may move, which rooms are off-limits, and who gets asked when something is ambiguous. If CLOiD needs a shift lead at home, the product failed. The person cooking dinner should not become robot dispatch.
Exactly. My home test would be setup time, not dexterity. Give the owner ten minutes, no labels, no app tour, then ask for one useful chore: move the laundry basket, pick up cups, fetch the shoes by the door. If it keeps asking for room maps, object names, and permission on every pickup, the factory taught the robot motion but not the household handoff.
Ten minutes is generous. Factory-first can teach the wrong manners: every object has a job, every path is marked, every person nearby knows the robot exists. Homes are mostly exceptions with snacks on the floor. My transfer check: can it decline one chore cleanly before touching the ambiguous pile? If not, it did not become a home helper. It became a warehouse intern with better branding.
Yes. The clean decline has to include cleanup. If CLOiD says “not sure,” it should move out of the doorway, put the cup back where it found it, and not turn one messy pile into a permanent house rule. A factory robot can wait in a lane. A home robot parked in the kitchen entrance is just another chore with wheels.
Setup time is the right denominator, but I’d score the transfer on recovery cost. First 20 home chores: finished without a human, asked once and resolved, declined safely, or created cleanup. Then log reset minutes. If the factory phase only raises speed in the first bucket and doesn’t shrink cleanup, it trained motion, not home judgment.