The robot autonomously completed dozens of uninterrupted stair-climbing cycles.
A lot of engineering goes into making autonomy look simple
Here’s a 2 hour timelapse of F.03 repeatedly walking up and down stairs. Figure HQ is full of tests like this, each one helping push robots closer to fully autonomous systems
No gantry!
A lot of engineering goes into making autonomy look simple
Here’s a 2 hour timelapse of F.03 repeatedly walking up and down stairs. Figure HQ is full of tests like this, each one helping push robots closer to fully autonomous systems
@adcock_brett that's the problem right there - you're treating it like a video game and only doing the observer draw - you have to do the environmental retention in ADDITION to the observer draw
@GoingBallistic5 @Figure_robot Not smart enough yet
@adcock_brett "What is my purpose?", "To walk up and down stairs", "Oh God". 😅
But seriously, this a major part of how robots are going to learn how to do things: learning by doing. There are no shortcuts.
@adcock_brett how are humans able to do that without such rigorous testing ? I feel like something about the confidence and efficiency of learning things like this is missing ..might be something simple
@adcock_brett @Figure_robot Why does he take the stairs when there is a short-cut?
/s
@adcock_brett the tedious work, is important. neat stuff
@adcock_brett Given how prevalent steps are, maybe it's worth building a rig that can easily adjust step height and width. I'm guessing ~99.99% of steps are between 3"-12" tall and between 7"-24" wide.
@SirMrMeowmeow @adcock_brett It’s not. You should know better having been yapping my about machine learning for years
hmm disagree, unless we’re using “important” very narrowly. data curation matters, and can be quite tedious if not already pre-made. real-world evals matter. hardware wear/breakage matters. stairs in sim are not the same as stairs after hours of actual contact, calibration drift, weird edge cases, etc.
you gotta leave the sim eventually, not everything is glamorous :x
@adcock_brett like your statistical confidence , however you want to measure that ( beside part testing or something for durability ) should need 1000 crossings of the same stairs
@adcock_brett What’s my purpose?
Do 10 thousand reps of stairs
@adcock_brett This is wild. 2 hours of stair practice on repeat - the progress on these robots is crazy.
@adcock_brett Make the stair heights uneven, then give the robot two bags of groceries to carry.
@adcock_brett bootstrapping the recursion more robots -> better robots -> more robots
getting off zero is brutal but it looks like we are there now
@adcock_brett okay😒
@adcock_brett Blazing!
@adcock_brett What’s the ETA?
@chris_j_paxton They're just flexing
@adcock_brett @Figure_robot Last I heard, figure has done around a million hours of data training with these huminoids. That’s hard to grasp. Even for me.