Why robots still cannot do the dishes: contact, cost and generality
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.
Robots still cannot do the dishes because handling objects means making and breaking contact with the world, and contact is where today's control methods, sensing and training data are weakest. On top of that, a general-purpose robot is expensive to make reliable and has to sell in the thousands to survive. None of the five experts says home robots are impossible, but all describe a slower, harder road than the one language models took.
Russ Tedrake, an MIT professor who heads robotics research at the Toyota Research Institute, locates the problem in contact. Most robots touch the world only with feet or fingertips. Contact is discontinuous, so methods built for smooth systems break down, and the number of ways surfaces can touch explodes past anything that can be enumerated. Household objects make it worse: a shirt has no simple state, and a chopped carrot becomes a hundred pieces. He says the real challenge for a home robot is validation, gaining enough confidence to ship it into anybody's kitchen.
Robert Playter, chief executive of Boston Dynamics, adds the economics. He calls it the curse of a general-purpose robot that it is not perfect at any one thing. His company builds machines costing a hundred thousand dollars or more, which industry can afford and consumers will not. His survival rule: "If you only sell hundreds, you will commercially fail." The path to the home, he says, runs through industry and mass manufacturing first.
Marc Raibert, who founded Boston Dynamics, separates athletic intelligence from cognitive intelligence and says robot hands are nowhere near human dexterity after 50 years of work. One video of Atlas climbing three steps took 109 tries. Homes, he notes, are harder than woods, because indoors you cannot leave scuff marks or bang into walls.
Kate Darling of the MIT Media Lab says, "I think we overestimate what we can automate," especially in the physical realm. Fei-Fei Li, the Stanford computer scientist, explains why the chatbot recipe does not transfer: AI learns from patterns in very large datasets, and there is no internet-sized record of physical handling.
They disagree about the humanoid body, which Tedrake defends because the world is built for the human form and Darling calls overhyped. They differ on timing too. Tedrake expects kitchen helpers before long, Playter says a consumer product might come eventually, and Li thinks in decades. The article adds a caution: Playter and Raibert lead industrial robot businesses, so industry first is also their strategy. These conversations span several years, so check current results.
What to do with it: when you see a robot demonstration, ask how many attempts it took and whether failures were shown. Ask what the robot touches and with what. Separate talking from doing, since a fluent voice says nothing about the hands. Check whether a product is purpose-built or general purpose, and look at the price and numbers sold. Expect robots to reach homes through industry first. If you build robots, design for contact with softer surfaces and tactile sensing, and budget for testing and breakage. Where data for a physical task is scarce, plan for a person and a machine working together.