@steve
you say "how clunky and a bit crap it is compared to the stuff Boston Dynamics was doing years ago." --- While I agree in part, I have an explanation for it. I don't think the quality gap is due to Cronos' musculoskeletal system design nor motor selections nor tendon based actuation. I think these can reach Boston Dynamics level of function nearly - at least enough to reach all my stated goals. The only thing lacking is the vast amount of AI and control algorithm perfecting and motion planning and motion smoothing and acceleration speed bio-mimicry to give the movements a human look and feel. The strength is there though - maybe not Boston Dynamics level explosive strength, but enough to achieve my stated goals which are MEGA ambitious goals mind you and reaching human level strength. Boston Dynamics was at some crazy strengths before due to using hydraulics but this was EXTREMELY loud and they regularly had massive oil explosions all over their workshop when their robots fell and a line tore. This is completely unacceptable for household use. Their oil messes were so common they had to line the floor with special mats and never could really get it fully clean. It was like a dangerous gyser every time. I digress. Boston Dynamics achieved the strength by using a poor choice of actuation for indoor use and that's why they are abandoning that approach now. It was doomed from the outset to be useless for long term robotics actuation in the future. Hydraulics are a poor choice period. Fine for a outdoor only robot I guess. Super expensive. Super heavy. Anyways, with more time spent fine tuning Cronos software, I think it would have Boston Dynamics level performance if not even better in movement (minus the explosive power of hydraulics). Also, modern BLDC motors for Cronos would help immensely as would modern processing and AI techniques. So you are missing the potential here in your assessment IMO.
You say "It appears unable to stand on its own without a tripod support." --- that's not true. Here's a video of it standing on its own without tripod support:
Note though that he never finished the legs or walking etc but all the potential was and is there still its just not done is all.
You say "I think BD have set the bar, and the only robotics worth anything need to be as good or better, otherwise, what's the point?" --- agreed. Absolutely agreed. But like I said, I think Cronos can hit that level with more software added. Its just got enough software to do some simple demos of the hardware. It is not anywhere near complete software wise. Only just a small humble beginning that is already very cool.
You say "If I'd built and programmed the robot in those vids I'd be proud and thrilled, but I would never dream of offering one up as an exemplar, or something to aim for. It's years behind." --- I disagree. Maybe a motors upgrade and more sensors and a ton of coding work added and its up to date.
@ebastler
You say "That's what I thought. Moving in slow motion with slightly wobbly motion, able to register a ball when it's held to its face and is the only red thing in the room" --- that's because its using just simple minimal AI as proof of concept to just show it off a bit. It hasn't had years of AI development. In fact, its mostly off the shelf AI not even really custom stuff. AI wasn't a huge focus of Cronos and can always be added by a serious AI developer later if they were to build the open source Cronos and all the issues you mentioned could easily be addressed as the AI becomes sophisticated and advanced.
You say "When I read "realistic humanoid robot", I would expect that machine to catch a ball which I throw to it, taking a few quick steps to position itself..." --- the robot itself has the human form factor and similar to human movement. With better software it could do all you said and more. Not even better software just MORE software. Like way more lines of code. That's all it needs. It is just a little start to the code project is all. Think "Hello World". A beginning. Just because it isn't finished doesn't mean it can't have skin added, code added, and become the realistic robot you imagined. It is in route to that but not done yet. Minus the cyclops eye thing of course, that part is easily swapped out though.
You say "And looking at the mechanism, getting there is not just a matter of adding a faster CPU and a stronger motor." --- that and more AI is exactly all it needs. What else are you suggesting it needs? I think its almost there. Not sure what you see differently than me.
You said "he never mentioned what age of a human he has chosen as his role model. Lets start with late 90s maybe?" -- this guy is I think like late 40s. He starred in Robot Wars decades ago I think. He might even be early 50s. He has multiple doctorates in robotics and engineering I think. He's very legit. Very highly respected. An industry leader IMO. A trailblazer. A visionary. A genius.
@zero
You say "That's not how it works. What you're planning to do is what causes model collapse: training an AI on the output of other AIs." --- sure, this can cause model collapse for LLMs theoretically. However, I'm not using an LLM and it won't be relying on this sort of data as its sole source of data. It will be using this only supplementally. The cause of collapse of my AI is not known and will never happen. My AI will be trained much more like raising a child with the use of interaction with real people as one of its primary means as well as the normal input of misc online data and book data and forum data and social media data and chatroom data and YouTube video data etc. Data from other AI's is not really for large scale brute force training like a LLM does. It's for supplemental assistance just like I personally use LLMs to ask questions. Using LLM's for you or for me does not crash our models does it? No. We have robust models with fail-safes and chatgpt and similar LLM outputs assist us but we don't trust it blindly and use our own logic to vet the information we get from it, testing it to see if it seems logical and accurate to us. My AI using 3rd party AI can do the same vetting we do when we use 3rd party AI for assistance.
Home
Help
Search
Login
Register
