Figure 03, our 3rd generation humanoid robot

Post by blackoil » Thu, Oct 09, 2025, 5:29 PM UTC

kibwen wrote: post #115

ipnon wrote: post #98

Earlier quoted context omitted.

Now the question is if this is GPT-2 and we’re a decade away from autonomous androids given some scaling and tweaks, or if autonomous androids is just an extremely hard problem.

For LLMs, the input is text, and the output is text. By the time of GPT-2, the internet contained enough training data to make training an interesting LLM feasible (as judged by its ability to output convincing text). We are nowhere near the same for autonomous robots, and it's not even funny. To continue to use the internet as an analogy for LLMs, we are pre-DARPANET, pre-ASCII, pre-transistor. We don't even have th…

Post by poisonborz » Thu, Oct 09, 2025, 5:33 PM UTC

proee wrote: post #65

websiteapi wrote: post #20

I don't understand why this even has to charge at all. It makes sense for multiple reasons to give it 3 batteries that say have 1/3 of the capacity, and make at least 1, if not 2 or 3 capable of charging independently on a station. Then the robot would just go to its station and swap its own batteries. Why even have wireless charging at all? Or even a cable? Or even have it "charge"? Battery swapping seems to make wa…

It depends on the battery life. If the robot lasts all day, then charging at night via standing on a charging pad makes a lot of sense. Creating a removable battery pack adds extra weight and gives the designers less freedom to place the battery pack exactly where it needs to be in the robot frame, or distribute the cells across the frame in strategic locations. Also, the charge rate matters. If robot can charge to 8…

Post by sosodev » Thu, Oct 09, 2025, 5:35 PM UTC

NewUser76312 wrote: post #140

People comparing this to GPT-2 is very interesting. While it sounds like a nice analogy or even a good story to investors, the fundamentals are very different. To train GPT, all of the training data (the internet of text, scanned books, etc) had already existed, even before the GPT project began. Arguably, the compute required (for GPT-3) also already existed, even before GPT-2. The GPT project really just came down…

Post by daveguy » Thu, Oct 09, 2025, 5:40 PM UTC

pizzathyme wrote: post #132

HAL3000 wrote: post #78

All of the examples in videos are cherry picked. Go ask anyone working on humanoid robots today, almost everything you see here, if repeated 10 times, will enter failure mode because the happy path is so narrow. There should really be benchmarks where you invite robots from different companies, ask them beforehand about their capabilities, and then create an environment that is within those capabilities but was not u…

How does this square with the video where they showed it running continuously for an hour doing an actual Amazon package sorting job? https://www.youtube.com/watch?v=lkc2y0yb89U

Post by 4b11b4 » Thu, Oct 09, 2025, 5:51 PM UTC

pants2 wrote: post #30

4b11b4 wrote: post #24

Earlier quoted context omitted.

Swapping requires a lot more moving parts and an additional enclosure to house the battery, and the batteries need to be much more rugged, and now you need two of them. But a cable is a fair question.. you'd think it could plug itself in... Maybe that's a hint at the robots actual capabilities at this point... or, they didn't want to bet on the unpredictability of environments: what if there's something in the way of…

I believe Elon has said before re:Tesla factory that plugging in cords is one of the hardest things to automate.