Really cool project.
Some questions to ML/AI researchers here on HN:
- Are there any ML/AI papers out there that trained on this? E.g. training robots in virtual environments?
- What data sets are used today in the ML/AI space to train robots?
From the video: "Infinigen is 100% procedural. Math rules only. Zero AI. Just graphics."
.
Timestamp: https://youtu.be/6tgspeI-GHY?t=107
"Zero AI" - nice to end the year with something so refreshing!
He’s asking if anyone is training using data created from a system like this, not if this uses AI.
The question wasn't "does this use AI". It was, "does any AI use this".
Exactly, thanks.
Did some light digging and found some related papers:
- https://robogen-ai.github.io/
- https://genesis-embodied-ai.github.io/, related HN thread: https://news.ycombinator.com/item?id=42457213
- https://paperswithcode.com/datasets?q=robot&v=th&o=match
These use other data sets, however.
I'm pretty sure i saw a video from Nvidia research using generated environments for training robots etc.
Can't find it quickly though.
Nvidias Omniverse is one of the core things though
If you look up the nvidia keynote from 2024, omniverse, robotics, and digital twins are discussed at length [1]. The relevant parts are near timestamp: 1 hour 34 minutes
Hasn't this been posted before? It's cool, I agree, but it's been around for a few years. Has there been an update?
Correct. 18-Jun-2023. 76 comments. https://news.ycombinator.com/item?id=36376071
The indoor scenes look completely plausible, but the outdoor leaves a lot to be desired. Very much behind the state of the art.
So what IS the state of the art that's also FOSS, perhaps even as-a-library, and non-gen-AI (I'm thinking interactive/game-dev/realtime use-cases)?
Not open source, but Gaia is widely used in games and movies. Can do some cool things like simulated vegetation growth/distribution.
You are probably thinking of gaea - not gaia. Gaia is also a world generator but not as widely recognised as gaea (and gaia is only implemented as a unity plugin IIRC).
Behind several versions of No Man's Sky, let alone whatever madness the LLM people have been cooking.
NMS isn't open source nor for the same purpose
You have to admit the sea water was way better than any video game today. The rest though, yeah.
What's the state of the art?
The fact that this is possible feels like it says something about the universe— that there is not as much information in our environment as we think there is.
This in turn could explain the unreasonable effectiveness of brains — especially small ones like animal brains — at modeling and operating in the world.
That's a very superficial way of looking at the universe.
Existence is not limited to whether or not it's observed. Brains have evolved to only gather the minimum amount of information required to keep the organism alive in its particular environment. But there is an unfathomable amount of complexity in the universe, even in the things we _can_ perceive.
For 3D modeling specifically these details don't matter, which is why we're able to approach photorealism. The renders look good enough to fool our brains into thinking that we're looking at the object as it would appear in the real world. They're a good approximation of visual aspects, but they're far from a good representation of real world objects. We don't have nearly enough compute for that. We've only recently gained the ability to model how light behaves in the real world in real-time, and even that is an approximation.
Anyway, Infinigen looks like a cool product. It's great to see classical simulations instead of AI for a change.
>The fact that this is possible feels like it says something about the universe
Similarly, the character models in Final Fantasy 7 say something about the human endocrine system. Aeris clearly reads as human despite you not being able to see her adrenal glands so maybe they don’t exist in real life
> "The fact that this is possible feels like it says something about the universe"
the map is not the territory
"There's plenty of room at the bottom."[0]
0. http://researchgroups.pku.edu.cn/_resources/group1/M00/00/0D...
I think it says more about human brains, how easily our perceptions are fooled.
The information density is large, our perception (i.e. the universe perceiving itself) filters for our particular length and time scale.
Sorry, the universe is abundant. This was the final thing Paul Fayerband was working on before he died…
it's unclear from the readme - is this infinite as in can generate infinite variations on a scene, or infinite as in generates an infinite world?
Previously : https://news.ycombinator.com/item?id=36373494
your meta tag descriptions need to be updated and are showing placeholder content when shared
I noticed several prominent Chinese universities in the authors list. Doesnt this prima facie imply a breach of the chip sanctions/export rules?
The papers mention that the co-authors did research while affiliated with Princeton.
Not even subtle with that Matrix-like intro
That is what they were going for, yes.