3 ms·

Why do people jump all the way to humanoid? That little box with an arm on it would not cost more than 10k, and I'd gladly take one for upstairs and one for dow

Why do people jump all the way to humanoid? That little box with an arm on it would not cost more than 10k, and I'd gladly take one for upstairs and one for downstairs WAY BEFORE I'd buy a 100kg walking steel frame to coexist with my toddlers.

falcor84 2y ago
The reason I jumped to humanoids is that I think that the fact that they'll be able to learn tasks by imitating us will make them a lot more versatile and cost-effective overall.
- actionfromafar 2y ago
  Isn’t current training paradigms requiring massive amounts of data? You aren’t going to get that easily from a handful of luxury robots.
  - corysama 2y ago
    Check out https://www.1x.tech/androids/neo https://www.1x.tech/androids/neo Their plan is to put droids in real homes and remote control them to create the kinds of training data that can’t be faked in a lab. They explain how that works here https://youtu.be/2ccPTpDq05A https://youtu.be/2ccPTpDq05A
  - falcor84 2y ago
    Why handful? I would assume that domestic robots like that would sell at least as many units as luxury cars - hundreds of thousands a year. In terms of training, there are a lot of current approaches of learning from videos (of which we have a lot) and from synthetic data, which is getting easier to generate. And if they need more data, I would assume that offering a version of such a robot with telemetry at a large discount would be an easy sell for some. In particular, I'd expect there would be a lot of demand for assistants to elderly people.
- bluGill 2y ago
  I'm not sure I want that. I would probably trade some less versatile robots that are cheaper and do 90% of the work. I can't afford a million dollars - or even 100k dollars for a robot. Until you get into the $1000 range it is hard to see the value. (maybe $20k at the top end - car makers will hate this though as you are directly competing with cars as the only way most people can get that kind of money is to keep their current car longer)
- jvanderbot 2y ago
  I'm not sure that the cost to build a near-flawless, learn-anything-by-imitation humanoid would be less than the cost of a "laundromatic bot" that weights 5kg and picks up socks and runs on 4 wheels. It just goes against all logic to say that a more general, significantly more capable system would be cheaper than a more-specifically-designed system.
  - falcor84 2y ago
    Obviously I'm not arguing that a humanoid will be cheaper than a laundromatic bot, but when I'm considering the full range of domestic tasks, I think it will start to make sense. Just for laundry, that would include: picking up items of all sizes, running the laundry, folding the clothes and then putting them in the closet. Other things I'd want help with include tidying (the defrag in the post), organizing my fridge&pantry and throwing away out of date items, cooking, vacuuming and mopping, and dozens of tasks that I'm sure would come to mind if I think for a bit longer. Essentially, I want a full domestic helper like Rosey from the Jetsons. I think that technologically we're getting quite close to that, with R&D examples like Atlas[0], Optimus[1], Figure[2], Mentee[3] and Neo[4]. If you're sure you'd be happy with just a laundromatic bot, that's cool, but I think that any non-humanoid approaches to domestic work are pretty much a dead-end, as they are the robotic equivalent of a non-programmable computer. [0] https://bostondynamics.com/atlas/ https://bostondynamics.com/atlas/ [1] https://www.tesla.com/AI https://www.tesla.com/AI (they apparently don't have a dedicated page) [2] https://www.figure.ai/ https://www.figure.ai/ [3] https://www.menteebot.com/bot/ https://www.menteebot.com/bot/ [4] https://www.1x.tech/androids/neo https://www.1x.tech/androids/neo

falcor84 2y ago

The reason I jumped to humanoids is that I think that the fact that they'll be able to learn tasks by imitating us will make them a lot more versatile and cost-effective overall.

actionfromafar 2y ago
Isn’t current training paradigms requiring massive amounts of data? You aren’t going to get that easily from a handful of luxury robots.
- corysama 2y ago
  Check out https://www.1x.tech/androids/neo https://www.1x.tech/androids/neo Their plan is to put droids in real homes and remote control them to create the kinds of training data that can’t be faked in a lab. They explain how that works here https://youtu.be/2ccPTpDq05A https://youtu.be/2ccPTpDq05A
- falcor84 2y ago
  Why handful? I would assume that domestic robots like that would sell at least as many units as luxury cars - hundreds of thousands a year. In terms of training, there are a lot of current approaches of learning from videos (of which we have a lot) and from synthetic data, which is getting easier to generate. And if they need more data, I would assume that offering a version of such a robot with telemetry at a large discount would be an easy sell for some. In particular, I'd expect there would be a lot of demand for assistants to elderly people.
bluGill 2y ago
I'm not sure I want that. I would probably trade some less versatile robots that are cheaper and do 90% of the work. I can't afford a million dollars - or even 100k dollars for a robot. Until you get into the $1000 range it is hard to see the value. (maybe $20k at the top end - car makers will hate this though as you are directly competing with cars as the only way most people can get that kind of money is to keep their current car longer)
jvanderbot 2y ago
I'm not sure that the cost to build a near-flawless, learn-anything-by-imitation humanoid would be less than the cost of a "laundromatic bot" that weights 5kg and picks up socks and runs on 4 wheels. It just goes against all logic to say that a more general, significantly more capable system would be cheaper than a more-specifically-designed system.
- falcor84 2y ago
  Obviously I'm not arguing that a humanoid will be cheaper than a laundromatic bot, but when I'm considering the full range of domestic tasks, I think it will start to make sense. Just for laundry, that would include: picking up items of all sizes, running the laundry, folding the clothes and then putting them in the closet. Other things I'd want help with include tidying (the defrag in the post), organizing my fridge&pantry and throwing away out of date items, cooking, vacuuming and mopping, and dozens of tasks that I'm sure would come to mind if I think for a bit longer. Essentially, I want a full domestic helper like Rosey from the Jetsons. I think that technologically we're getting quite close to that, with R&D examples like Atlas[0], Optimus[1], Figure[2], Mentee[3] and Neo[4]. If you're sure you'd be happy with just a laundromatic bot, that's cool, but I think that any non-humanoid approaches to domestic work are pretty much a dead-end, as they are the robotic equivalent of a non-programmable computer. [0] https://bostondynamics.com/atlas/ https://bostondynamics.com/atlas/ [1] https://www.tesla.com/AI https://www.tesla.com/AI (they apparently don't have a dedicated page) [2] https://www.figure.ai/ https://www.figure.ai/ [3] https://www.menteebot.com/bot/ https://www.menteebot.com/bot/ [4] https://www.1x.tech/androids/neo https://www.1x.tech/androids/neo