This seems so reminiscent of the early 1900s moment when at first people thought the way to electrify a factory would be to have a big electric motor to power all the belt drives instead of a steam engine.
Yeah I think that's spot on -- you need to build entirely new systems to take advantage of many new technologies instead of directly swapping things out.
The cost scaling of microprocessors is one way to think about it and I would be surprised if some who expect drop-in humanoid robots weren't thinking about LLMs and how they're increasingly used in place of specially-trained AI. It would be interesting to read your reasoning why that comparison doesn't make sense (and even more so if you think it does!)
Would it go something like "a few companies are racing each other to bring humanoid robots to market and capture the biggest share of the market, so they raise billions from Softbank etc and sell at a loss, bringing the prices down below specialized hardware?"
I think robotics companies spend a ton on algorithms, and need to amortize that cost over their small volumes. But that’s not a marginal cost, so if volumes were to go way up, or if open-source algorithms were to get better, then we would want to ignore the part of the cost related to algorithm R&D. This leads into the idea that maybe we should be looking at the prices of teleoperated robots instead of autonomous ones. (The teleoperation UI takes R&D too, so it’s still an overestimate, but probably by less.) So, I tried to look into teleoperated robot prices a year or two ago. One thing I found was the "Ugo" — https://spectrum.ieee.org/remotely-operated-home-robot-can-do-your-laundry — which might or might not be vaporware, but they mention a price point below $10/day. I also found the much more hardcore "Sarcos Guardian XT" — https://www.youtube.com/watch?v=4urGVpy5HiA — which is possibly discontinued and never had transparent pricing, but I found a site that mentioned that you lease it for $5K/month. (And it does have dexterous hands, IIUC.) That's comparable to your higher-end numbers, but I think it should be rounded way down because I imagine that they were selling tiny volumes, and I think also designing them for great strength and ruggedness in extreme environments (e.g. high temperatures).
Also, individual humans can make pretty cool mechanical hands! https://www.youtube.com/watch?v=kixS5Sy0F9s . I doubt the guy spent more than $1000 on those parts, right? (Could be wrong.) We can chat about the extent to which this is relevant / analogous—I’m not too sure, I’m kinda confused about what that guy did exactly and how it works. Whatever it is, it looks cool though!
This seems so reminiscent of the early 1900s moment when at first people thought the way to electrify a factory would be to have a big electric motor to power all the belt drives instead of a steam engine.
Yeah I think that's spot on -- you need to build entirely new systems to take advantage of many new technologies instead of directly swapping things out.
The cost scaling of microprocessors is one way to think about it and I would be surprised if some who expect drop-in humanoid robots weren't thinking about LLMs and how they're increasingly used in place of specially-trained AI. It would be interesting to read your reasoning why that comparison doesn't make sense (and even more so if you think it does!)
Would it go something like "a few companies are racing each other to bring humanoid robots to market and capture the biggest share of the market, so they raise billions from Softbank etc and sell at a loss, bringing the prices down below specialized hardware?"
I think robotics companies spend a ton on algorithms, and need to amortize that cost over their small volumes. But that’s not a marginal cost, so if volumes were to go way up, or if open-source algorithms were to get better, then we would want to ignore the part of the cost related to algorithm R&D. This leads into the idea that maybe we should be looking at the prices of teleoperated robots instead of autonomous ones. (The teleoperation UI takes R&D too, so it’s still an overestimate, but probably by less.) So, I tried to look into teleoperated robot prices a year or two ago. One thing I found was the "Ugo" — https://spectrum.ieee.org/remotely-operated-home-robot-can-do-your-laundry — which might or might not be vaporware, but they mention a price point below $10/day. I also found the much more hardcore "Sarcos Guardian XT" — https://www.youtube.com/watch?v=4urGVpy5HiA — which is possibly discontinued and never had transparent pricing, but I found a site that mentioned that you lease it for $5K/month. (And it does have dexterous hands, IIUC.) That's comparable to your higher-end numbers, but I think it should be rounded way down because I imagine that they were selling tiny volumes, and I think also designing them for great strength and ruggedness in extreme environments (e.g. high temperatures).
Also, individual humans can make pretty cool mechanical hands! https://www.youtube.com/watch?v=kixS5Sy0F9s . I doubt the guy spent more than $1000 on those parts, right? (Could be wrong.) We can chat about the extent to which this is relevant / analogous—I’m not too sure, I’m kinda confused about what that guy did exactly and how it works. Whatever it is, it looks cool though!