[HN Gopher] Physical Intelligence's first generalist robotic model
___________________________________________________________________
Physical Intelligence's first generalist robotic model
Author : lachyg
Score : 56 points
Date : 2024-10-31 21:34 UTC (1 hours ago)
(HTM) web link (www.physicalintelligence.company)
(TXT) w3m dump (www.physicalintelligence.company)
| cryptoz wrote:
| At 1:50, the guy gives the robot a glass to pick up and then
| immediately nopes out of there. Wonder if previous demos resulted
| in a broken glass haha.
|
| Also at 2:08 the upside-down container gets flipped quickly. I
| wonder if that was a known limitation of the robot at the time or
| if the person just had a desire to flip it right-side up (to be
| polite? haha).
|
| I'm commenting on these tiny details and laughing a lot because
| I'm not sure I can handle a more serious approach to this.
| Doesn't it seem like in < 10 years there will be dozens of
| autonomous, affordable home-robots? Everything is going to
| change.
|
| One last note, they call this generalist, but each of the
| examples is quite specific from a macro perspective. Yes the
| robot can fold maybe any pile of crumpled laundry now and that is
| generalist compared to previous efforts, but seems like we
| shouldn't be trying to train bots how to do billions of tasks in
| specific detail; rather they should learn to learn and take on
| new tasks they weren't trained for.
| OrigamiPastrami wrote:
| > Doesn't it seem like in < 10 years there will be dozens of
| autonomous, affordable home-robots?
|
| If you buy the hype, sure. I know many startups that have
| already gone bust working on this. I've also seen lots of
| similar attempts in laboratories around the world going back
| well over a decade.
|
| > One last note, they call this generalist, but each of the
| examples is quite specific from a macro perspective. Yes the
| robot can fold maybe any pile of crumpled laundry now and that
| is generalist compared to previous efforts, but it does seem
| like we shouldn't try to train bots how to do billions of tasks
| in specific detail; rather they should learn to learn and take
| on new tasks they weren't trained for.
|
| You are starting to see how difficult the problem is and how
| limited the solutions are. You're basically saying "let's just
| give the robots general AI and everything will be so much
| easier!"
| golol wrote:
| Idk this is really promising, how many robot foundation
| models have you seen before that also work very well? I
| believe this is all quite recent.
| OrigamiPastrami wrote:
| I'm not saying there isn't progress. I'm saying progress is
| slow relative to the work that needs to be done. I've also
| worked at enough robotics companies to be skeptical of
| anything they publish because there is a _strong_ tendency
| to cherry-pick results. The disconnect between the research
| papers being published and the reality of the robots at one
| company I worked at was pretty egregious.
|
| Robots are super cool. Just be skeptical of the hype.
| edm0nd wrote:
| I think it would be super awesome. I hate doing laundry so if
| someone sold a robot that washed + dry + folded all my laundry,
| I would spend money on it.
|
| I'm talking about I want to throw my dirty clothes into a
| basket and it takes care of the rest.
|
| The demo from the video gives me hope!
| golol wrote:
| This is a duplicate thread. Can some mod merge them oO? I don't
| know how this works on HN.
| lachyg wrote:
| (I work at p.)
|
| Happy to answer any questions on the model, hardware, etc
| imranhou wrote:
| First of all - incredible work. Do you guys plan to integrate
| frameworks like ROS to help manage this robot?
| golol wrote:
| I saw your foundation model is trained on data from several
| different robots. Is the plan to eventually train a foundation
| model that can control any robot zero shot? That is, the effect
| of actuations on video/sensor input is collected and understood
| in-context and actuations are corrected to yield intended
| behavior. All in-context. Is this feasible?
|
| More specifically, has your model already exhibited this type
| of capability, in principle?
| dr_dshiv wrote:
| Nearly 2 years ago I bet a roboticist $10 that we'd have "sci-
| fi" robots in 2 years.
|
| Now, we didn't set good criteria for the bet (it was late at
| night). However, my personal criteria for "scifi" are twofold:
| 1. Robots that are able to make peanut butter sandwiches
| without explicit training 2. Robots able to walk on sand (eg
| Tatooine)
|
| Based on your current understanding, who won the bet? Also,
| what kind of physical benchmarks do you associate with "sci-fi
| robots"?
| owenpalmer wrote:
| At 2:54, it struggles to pick up the cloth for 10 seconds (100
| seconds real-time).
|
| This may just be a software fix, but I wonder about the idea of
| exchanging tools for different tasks. In this case some kind of
| pincher-vacuum or roller-grip might have done the job better.
| yalogin wrote:
| This is actually promising. I hope these guys continue to iterate
| for how much ever time they need to
___________________________________________________________________
(page generated 2024-10-31 23:00 UTC)