Why robots still cannot do the dishes: contact, cost and generality
October 5, 2026 · 14 min read
Russ Tedrake, Robert Playter, Marc Raibert and Kate Darling explain that touching and handling the physical world remains far harder than language, and that general-purpose machines are expensive to make reliable. It tempers predictions that household robots will follow quickly behind chatbots.
The rules, numbers and steps worth keeping in front of you.
- Core reason: manipulation means making and breaking contact, where control, sensing and data are weakest.
- Tedrake: contact is discontinuous, and the combinations of which surfaces touch when cannot be enumerated.
- Tedrake: deformable and divisible objects, like a shirt or chopped carrots, have no simple state to track.
- Tedrake: the real challenge for a home robot is validation, enough confidence to ship it into any kitchen.
- Tedrake: soft hands turn point contacts into patch contacts, so a hand can cradle an egg.
- Playter: a general-purpose robot is perfect at nothing; a purpose-built machine costs less and fails less.
- Playter: a robot maker that sells only hundreds will commercially fail.
- Playter: the path to the home runs through industry and mass manufacturing first.
- Raibert: one Atlas stair-climb video took 109 tries over about six weeks.
- Raibert: homes are harder than woods, because indoors you cannot scuff or bang into walls.
- Darling: we overestimate what we can automate, especially physically.
- Li: there is no internet-sized dataset of physical handling, so the chatbot recipe does not transfer.
- Ask of any demo: how many attempts, and what does the robot actually touch?
- A fluent voice on a robot says nothing about its hands.
This is the short form. The full insight has every expert's reasoning, their own words and the sources.