New Physics Sim Trains Robots 430,000 Times Faster Than Reality (arstechnica.com) 26
An anonymous reader quotes a report from Ars Technica: On Thursday, a large group of university and private industry researchers unveiled Genesis, a new open source computer simulation system that lets robots practice tasks in simulated reality 430,000 times faster than in the real world. Researchers also plan to introduce an AI agent to generate 3D physics simulations from text prompts. The accelerated simulation means a neural network for piloting robots can spend the virtual equivalent of decades learning to pick up objects, walk, or manipulate tools during just hours of real computer time.
"One hour of compute time gives a robot 10 years of training experience. That's how Neo was able to learn martial arts in a blink of an eye in the Matrix Dojo," wrote Genesis paper co-author Jim Fan on X, who says he played a "minor part" in the research. Fan has previously worked on several robotics simulation projects for Nvidia. [...] The team also announced they are working on the ability to generate what it calls "4D dynamic worlds" -- perhaps using "4D" because they can simulate a 3D world in motion over time. The system will reportedly use vision-language models (VLMs) to generate complete virtual environments from text descriptions (similar to "prompts" in other AI models), utilizing Genesis's own simulation infrastructure APIs to create the worlds.
"One hour of compute time gives a robot 10 years of training experience. That's how Neo was able to learn martial arts in a blink of an eye in the Matrix Dojo," wrote Genesis paper co-author Jim Fan on X, who says he played a "minor part" in the research. Fan has previously worked on several robotics simulation projects for Nvidia. [...] The team also announced they are working on the ability to generate what it calls "4D dynamic worlds" -- perhaps using "4D" because they can simulate a 3D world in motion over time. The system will reportedly use vision-language models (VLMs) to generate complete virtual environments from text descriptions (similar to "prompts" in other AI models), utilizing Genesis's own simulation infrastructure APIs to create the worlds.
No it doesn't (Score:3)
Re: (Score:3)
Actual physics, depending on the materials and forces involved, can easily get down to atomic-level modeling. You can treat much of that statistically to keep it macroscopic, but that means lots of special case rules.
If you want to ask "what happens when when I bang two rocks together" and are OK with a simple stress analysis, a loud noise, and a probability of one or both breaking... fine. If you want to ask what happens when marcasite and flint strike each other in various ways, things get vastly more c
Re: (Score:2)
Uhm, ok, sure.
Re: No it doesn't (Score:2)
Re: (Score:3, Funny)
Re: (Score:2)
I don't think you're in any position to lecture others about the difference between reality and fiction.
Re: (Score:2)
Re: (Score:2)
I guess you chose the blue pill.
Re: (Score:2)
Neither this generation of Programmers, nor this generation of AI, nor the Robot bodies they are trying to train, understand the difference between reality and fiction.
GOOD NEWS FRIENDS! (Score:2)
The robots will just wind up stuck in a corner (Score:2)
Journalism at Its Finest (Score:2)
The team also announced they are working on the ability to generate what it calls "4D dynamic worlds"—perhaps using "4D" because they can simulate a 3D world in motion over time.
If only there were someway the journalist writing the article could have actually found out why they named it 4D, you know by perhaps asking the term? How are we supposed to take an article like this seriously if the journalist writing it is speculating about what they are writing about instead of doing their job, finding out and then reporting it.
Indeed I'd question whether they have really simulated time as a true fourth dimension because somehow I doubt any current robots need to account for relativ
News? (Score:2)
I was building a lab for a professor some time back. He was concerned about performance because he wanted to simulate financial markets in real-time. I asked him "why not faster?", He said it would provided the response that it had to be real time for accuracy. I asked "time stamped event queue?", he gave me a blank stare.
We all go faster... Except fools
Sounds legit! (Score:2)
The text input uses AI to create the layout for a 3D environment which models the real world perfectly with no clipping and accounts for all the fluid dynamics, vibration resonance, bending loads and production tolerances etc. This is then run millions of times with no human supervision and produces reliable results.
something fishy (Score:2)
Why does it make sense to train something with 10^14 iterations of a known algorithm when you already know the algorithm?
Wow all that from a cell phone sim card! (Score:2)
Impressive!