01:07:55 · Michael Cho, Vincent Liu, Chris Paxton
[01:07:55] S0: Yeah. I think the... Like, at that point about, like, just keeping... I think when he says keep things simple, meaning keeping it simple ergonomically for the human. Yeah. Because the... I mean, we're in a phase where we just need to bootstrap learning from humans, like, for the robots to even start to work Yeah. And be somewhat autonomous so they can start to learn on its own. And so you really want to kind of, like... It's it's it's same line of, like, thinking with projects like UMI, you know, text UMI, text Wow, and and this is, like, to the other extreme. Right? Basically, nothing, just your glasses and just click record, that's it. And so, yeah, you can... I can I can totally see it being also very, very diverse set, and and also the scale that you can collect is is gonna be way more? I mean, in theory, the simpler it is. So... Yeah. Especially if this hardware keep going down in price, then a lot more people would just just have it. And and so I think that helps a lot. Are you familiar with Ego4D and and, you know, Ego Kitchen? Yeah. Yeah. I suppose... Yeah. Maybe I should have asked them just now, but I I imagine it's it's along the same kind of, like, angle trying to... [01:09:12] S1: You know? [01:09:13] S3: Yeah. I mean, yeah, that's true. But, honestly, the other thing is, like, a lot of that data... Like, if it's not collected for robots, it's, like, some... So so far, I think those haven't been quite as useful. It's my... It's been my impression. But, like...
01:09:13 · Chris Paxton, Michael Cho
[01:09:13] S3: Yeah. I mean, yeah, that's true. But, honestly, the other thing is, like, a lot of that data... Like, if it's not collected for robots, it's, like, some... So so far, I think those haven't been quite as useful. It's my... It's been my impression. But, like... [01:09:28] S0: Interesting. When you say collected for robots, you literally mean maybe being very specific to the human. Okay. The way you do it, make sure you show your head and do it like like that. Like... You know? [01:09:39] S3: Yeah. Like, a lot of the Ego four d videos, for example, like, if it's a person performing a task at person speeds, right, like, the cameras. This is also earlier versions of Aria. Right? So, like, the cameras would blur, and you wouldn't be able to see, like, what was going on and, like, all the features would get washed out. This is, like... Yeah. There's still, like, a lot of ways to go, I think, as far as video models to... Before we can get that kind of stuff to work well. Yeah. Yeah.
01:09:39 · Chris Paxton, Michael Cho, Vincent Liu
[01:09:39] S3: Yeah. Like, a lot of the Ego four d videos, for example, like, if it's a person performing a task at person speeds, right, like, the cameras. This is also earlier versions of Aria. Right? So, like, the cameras would blur, and you wouldn't be able to see, like, what was going on and, like, all the features would get washed out. This is, like... Yeah. There's still, like, a lot of ways to go, I think, as far as video models to... Before we can get that kind of stuff to work well. Yeah. Yeah. [01:10:02] S0: But But I suppose having hardware, like, built in that can help... Really help you with the SLAM. So bare minimum, you have decent camera pose. Yeah. Got it. Mhmm. I think that already helps. So you... You're not always just relying on computer vision. Because I can imagine you just shake your head like that, and then the whole image is just, like, almost not reducible. It's just, like, blurry images. Yeah. Whereas if you have a very decent IMU here, maybe you you can recover that a lot more. K. Yeah. But I I... Yeah. I'll be keen to see if if there... Yeah. If there are more papers along this line, let... Let's bring them Yeah. Bring them on. Yeah. Because it sounds like these guys are gonna do two things in in... As a company now, so maybe I don't know whether they will still publish it. [01:10:48] S3: Maybe. At least see it. [01:10:50] S0: Yeah. We'll see. [01:10:50] S3: There's lots of Yeah. There's lots of exciting stuff happening in that world right now. Yeah. [01:10:55] S1: Very cool. Yeah. For
01:10:55 · Vincent Liu, Michael Cho, Ademi Adeniji, Chris Paxton
[01:10:55] S1: Very cool. Yeah. For [01:10:56] S0: sure. Alright. Cool. Alright. I guess until the next one. [01:10:59] S2: Yeah. See you [01:10:59] S3: next time. Bye. [01:11:00] S0: Chat soon.