Robots struggle with dishes because real-world manipulation, safety, cost, and data lag behind human-evolved dexterity.
Why can’t a robot do the dishes? Well, it can, but maybe not as well or as cheaply as we’d hoped. More than once, I have started a robotics article with an appeal to the Jetsons tv show of the late twentieth century, and as we lament “where are our flying cars?” so, too, do we lament the lack of a robot butler or maid who, like the aproned helper in the cartoon, can do those household chores that we consider tedious and toilsome.
But now, as we actually have a lot of this type of technology coming online for the first time, we sort of have a clearer picture of how difficult it is. Here are some things that I’ve been hearing about the limitations of robotic AI, things that make a lot of sense considering the slow advent of the dishwashing automaton.I recently came across this for the first time, researching why it is that, for example, Deep Blue beat Kasparov in 1997 but we still don’t have robots doing the chores.states that although computers can walk all over us in terms of pure intelligence, doing things in the real world is still the domain of the human. You only have to scratch the surface to see that notables like our own Marvon Minsky have been familiar with this concept since the 1980s when it was attributed to Hans Moravec, an Austrian who worked at Carnegie Mellon. To me, though, it’s the reasoning behind this that’s so fascinating. Why? Why are humans so good at manipulating objects in the meatspace, and why is it so hard for robots to follow suit? The answer? Charles Darwin. Or, to be accurate, the principle of natural selection. This argues that we became better at these things over time – over, say, a few million years.Do you think that proto-humans were less dexterous, or is it more of an academic theory?Either way, Moravec’s paradox goes a long way toward explaining why it’s going to take more than sensors for robots to really be effective in the real world. The challenge combines fine motor skills and decision making, knowledge of physics and intuition. As for knowledge of physics, a lot of that may be intuitive, instinctive, built in – to us. Isn’t that weird?Another way to see the limits of practical robotics is to get lists from LLMs. Microsoft Copilot, when prompted, simply mentions the following: complexity of humanoid robots, market challenges, lack of efficiency, technological limitations, and “investment and patience,” re-presenting the bogey of ROI that dooms so many peripheral technologies until their time once again comes around.A: “Because chores are a nasty combo of messy perception + tricky hands + endless edge cases. A robot has to recognize clutter in changing light, plan safe motions around pets/kids, and manipulate squishy, slippery, crinkly stuff . Homes also aren’t standardized. Humans generalize effortlessly; robots still struggle with that robustness at low cost.” Well, yes, you have to handle trash bags. And sponges. But I think that other part, “plan safe motions around pets/kids” is a larger piece of the puzzle.where Navid Aghasadeghi, someone who has worked on these projects for Boston Dynamics and others, notes the various hurdles to building chore robots. “We don't have robots like Rosie from the Jetsons that can go around the house, do the dishes, do the laundry, so you can spend some more time with your family,” Aghasadeghi said. “And I should say that I feel partially at least responsible for this, because I've been working in robotics for the past 15 years.”“You have to get the software right,” Aghasadeghi added. “That's the intelligence. The software has to tell the robot how to interpret the world, and it also has to tell the robot what sequence of actions to take to actually do something useful. Once you have that, you obviously need the hardware. So you need actuators that can move the joints of the robot and as a whole, can move the entire body of the robot, and have the robot exert forces on the environment and do useful things.” Then, too, he continued, you need a lot of sensors to help the robot to accurately perceive the world around it, and you need to add all of these things at scale.“We get to build a chess engine that can beat a grand master of chess before we can build robots that can take simple steps, or before we can build robots that can grasp objects of different sizes, manipulate them, and stack objects together,” he said. “These are things that a one or two year old human baby learns to do very fast, and robots have been failing at this for a long time, and the internet is filled with a lot of videos of robots failing up until now.”To explain, Aghasadeghi evoked three distinct phases of robotics, up to the present day.“These were robots that would go from position A to position B,” Aghasadeghi said. “They didn't really have a perception, so they couldn't really see the environment. They just knew how to execute this task. And because of this, they were inherently not safe, and they could operate in only environments that were precisely manufactured and everything was in the right place.” The second stage he identified was “planning and perception,” where robots started to be able to see and understand the world around them, and reason about multi-stage tasks. The third is what he called “AI-powered robots,” where the AI in the robots imbues them with a kind of uncanny intelligence. “These robots are powered by the same kind of AIs that drive your ChatGPT and drive your image generation technology, and these are technologies that actually generalize beyond tasks,” he said. “So essentially, we have general learning machines now, and general learning algorithms … the same LLM that is driving your ChatGPT … is now planted into a robot brain. And someone can ask the robot to clean their kitchen, close the cabinet and do the dishes, and the robot is able to first understand the language, understand the prompt plan that, as a series of actions, it has to do to achieve the task, and then also understand the environment, and as a function of that, decides on these series of tasks.”“Does this mean that robotics is solved?” Aghasadeghi asked rhetorically. “The answer is, not quite yet, and the main problem … is the problem of data.” In short, Aghasadeghi suggests that although ChatGPT can use the vast reaches of the web to learn about facts, the real-world data base is not there to empower robots to learn, for example, how to hold objects. He added that researchers are hard at work on the task of “teleoperation,” to build such data sets for a new generation of robots. “Does this mean that we're going to have robots running around and doing things for us next year?” Aghasadeghi said. “I would say we're not quite there yet. And the analogy here is that robots are where Palm Pilot was a few years ago, if you are old enough to remember that, and robotics needs the revolution that iPhone had to happen, for robotics to actually be able to be all around us … robotics can actually serve as a mirror to humanity. So through understanding what technologies we have to build to augment humans, we actually get a better understanding of ourselves, and that is the future I'm hoping for, to be able to build technologies that … let us flourish even more.” All of this is a fascinating foray into the hurdles of roboticized AI and suggestions on how to move boldly and confidently ahead. And someday, robots will be doing the dishes, so you don’t have to.
United States Latest News, United States Headlines
Similar News:You can also read news stories similar to this one that we have collected from other news sources.
Marine robotics firm will resume deep-sea search for MH370 plane that vanished a decade agoThe search will be carried out by Texas-based marine robotics firm Ocean Infinity, which signed a new “no-find, no-fee” contract with Malaysia's government in March.
Read more »
Marine robotics firm will resume deep-sea search for MH370 plane that vanished a decade agoOcean Infinity will conduct the deep-sea hunt under a “no-find, no-fee” contract with Malaysia’s government.
Read more »
Marine robotics firm will resume deep-sea search for MH370 plane that vanished a decade agoMalaysia’s transport ministry said Wednesday that a private firm will resume a deep-sea hunt for Malaysia Airlines Flight 370 later this month, more than a decade after the jet vanished without a trace.
Read more »
Vin Diesel’s Rock ‘Em Sock ‘Em Robots Movie Gets Exciting Update After 4 YearsThe Rock 'Em Sock 'Em Robots live-action movie received an exciting update involving star and producer Vin Diesel.
Read more »
Bio-design leap: Food waste becomes functional robotic hardware in new researchResearchers at EPFL create robotic devices from langoustine exoskeletons, showcasing soft robotics and nature-driven engineering.
Read more »
Humanoid robot farmer? Malaysia plans to integrate bots to advance vertical farmingA Malaysian agricultural technology firm has partnered with UBTECH Robotics, a global leader in humanoid robots, to debut its new initiative.
Read more »