I’m highly entertained by the way the robot tightens nuts. There is no way this would work in the real world :)
(This whole thing is awesome. I think they will need to get the number of rollouts needed down before it translates well to real robots — running hundreds of thousands of rollouts on small parts with a real rollout will be a real logistical mess when cleaning up parts on the floor and dealing with everything that the robots physically break while learning.)
We use sim2real with distillation-based transfer :)
This means no need to do the rollouts on real robots first! Please check out https://sgs-rl.github.io for some transfer examples (albeit a simpler setting then the multi-part NIST board we examine in AlphaNIST).
I know you are joking, but that is absolutely a possible trajectory. Reliable and cheap LLMs could lead to software that re-implements everything, instead of using libraries.
The starting states being at different points in the task and having success-guided starting state sampling seem like really neat ideas. I haven't kept up with RL in a while, how novel is this?
The mechanism is loosely related to curriculum learning, though we argue there are some distinctions. In the methodological companion paper (https://sgs-rl.github.io), we try to get into more details in the related work and baselines including comparison to methods like PLR (also from Deepmind), SFL, and linear handcrafted curriculums over states.
You are giving me some inspiration, though, that it might be good to try to explain the related techniques in a lower-dimensional setting to help teach the concept more effectively! brb
We can answer questions! But also, I want to highlight that the "AlphaNIST" name is an homage to the very well known work from David Silver's research group @ Deepmind circa 2016->2018. This was when we were coming up very early in our research careers and their work obviously had a big impact on us!
It has taken longer to make some of those ideas work for areas as challenging as robotics where there are physical constraints to deal with (and admittedly in this paper, we do not show the transfer to real hardware quite yet). But, we have some glimmer of evidence it is looking possible here (in another recent paper) that we cranked on for the last 6 months and will be featured at CoRL: https://x.com/octi_zhang/status/2108583315544682914
We are also all mega-lurkers here so we were honored to be featured on the front page!
Point received! We wonder what a better way to pay that same homage would be, since we followed the similar philosophy, but with considerations for our contact-rich robotic control setting.
As far as I know Apple also does not own "i" or McDonalds owns "Mc".
If you're creating a Burger called McSomething or a device called iWhatever it does not feel like "paying homage" to Apple/McDonalds, it feels more like copying and trying to capture some audience from this bigger player.
Ah, I see, Ireland and McKinsey are trying to copy and capture McDonalds audience. /s
Yes, you said burger, but also AlphaNIST is not a technology conglomerate holding company. And the term Alpha is not at all used in their products. It's neither Alpha Phones, not AlphaGemini, nor is Google called Alpha. More than that Alphabet is never shortened Alpha.
I would strongly argue against that. I would argue that Google Deepmind has consistently used AlphaSomething for their scientific models. Not for their consumer models which are Gemini, but their models which are more cutting edge and specific to niche problems.
Got any in a single category which all start with Alpha and named by the same organization? Say something like... I don't know... Deep learning models published by Google?
If there's 10 breakfast cereals from the same company all starting with Alpha, I would give you some credit. But like this I don't even know what your reply is trying to tell me.
I honestly feel like I am taking crazy pills here.
OK I'm telling you that I'm not as keen as you are to cede every possible name that starts with "Alpha" to Google.
And to answer your question directly, DEC had a whole series of things called "Alpha". eg AlphaStation family of workstations, the Alpha AXP series of processors etc (alpha 21064, alpha 21164, alpha 21264, alpha 21364).
That's because Google Deepmind has consistently used AlphaSomething for their scientific models. Not for their consumer models which are Gemini, but their models which are more cutting edge and specific to niche problems.
Very cool line of work and emergent behaviors are so fascinating and solving NIST standard is very impressive! This just shows robot dexterity doesnt necessarily always need hands atleast for industrial assembly tasks.
Came across your CoRL publication before I found alphaNIST. I was curious how much simulation asset engineering and sysID/tuning was needed for transfer to the real robot?
Ensuring the assets are "CAD-level accurate" has a huge impact on making this work in practice. So much so, that our collaborators Yash Narang [1] and Iretiayo Akinola [2] have spent a number of years on just this - we are standing on the shoulders of giants!
Additionally, understanding the hardware well (including things like motor and controller dynamics) is non-negotiable. We learned the hard way that there is, in fact, no free lunch. You really have to get the deep understanding of physics and engineering right.
Really cool stuff. I especially liked the flicking behavior. Very organic movements.
In case somebody got the same question as me: the robot "sees" point clouds as input, so basically the full simulation state except for velocities (correct me if I'm wrong). As put in the limitations section, it "[does] not establish robustness to perception error, latency, or contact-model mismatch".
The movements of the robot gripper arm and its interaction with the longer objects ("Single-component assembly experiment", 1:00 and onwards) somehow really reminds me of how a horse moves it head.
Kuyawa | 22 hours ago
Alphanist - Some app/lib/code that does something
[OP] jasonjmcghee | 22 hours ago
HN guidelines are to use the title of the blog post and not editorialize.
jagged-chisel | 21 hours ago
[OP] jasonjmcghee | 21 hours ago
amluto | 22 hours ago
(This whole thing is awesome. I think they will need to get the number of rollouts needed down before it translates well to real robots — running hundreds of thousands of rollouts on small parts with a real rollout will be a real logistical mess when cleaning up parts on the floor and dealing with everything that the robots physically break while learning.)
romesc | 17 hours ago
This means no need to do the rollouts on real robots first! Please check out https://sgs-rl.github.io for some transfer examples (albeit a simpler setting then the multi-part NIST board we examine in AlphaNIST).
Mr_P | 22 hours ago
It is not an Alphabet/Deepmind paper, and it also has nothing to do with the "NIST" OCR dataset.
Was there really not a more appropriate name than "AlphaNIST"?
just2answer | 22 hours ago
[OP] jasonjmcghee | 22 hours ago
Alpha historically had meant "Google DeepMind" not RL
sdenton4 | 20 hours ago
Den_VR | 17 hours ago
Situation Soon: there are 15 trillion standards tailored to every possible situation.
grumbelbart2 | 12 hours ago
geerlingguy | 17 hours ago
Y_Y | 14 hours ago
wasabi991011 | 22 hours ago
romesc | 17 hours ago
You are giving me some inspiration, though, that it might be good to try to explain the related techniques in a lower-dimensional setting to help teach the concept more effectively! brb
romesc | 21 hours ago
We can answer questions! But also, I want to highlight that the "AlphaNIST" name is an homage to the very well known work from David Silver's research group @ Deepmind circa 2016->2018. This was when we were coming up very early in our research careers and their work obviously had a big impact on us!
It has taken longer to make some of those ideas work for areas as challenging as robotics where there are physical constraints to deal with (and admittedly in this paper, we do not show the transfer to real hardware quite yet). But, we have some glimmer of evidence it is looking possible here (in another recent paper) that we cranked on for the last 6 months and will be featured at CoRL: https://x.com/octi_zhang/status/2108583315544682914
We are also all mega-lurkers here so we were honored to be featured on the front page!
serioussecurity | 20 hours ago
romesc | 18 hours ago
dpe82 | 17 hours ago
marcbeep | 12 hours ago
jcattle | 12 hours ago
If you're creating a Burger called McSomething or a device called iWhatever it does not feel like "paying homage" to Apple/McDonalds, it feels more like copying and trying to capture some audience from this bigger player.
tete | 10 hours ago
Yes, you said burger, but also AlphaNIST is not a technology conglomerate holding company. And the term Alpha is not at all used in their products. It's neither Alpha Phones, not AlphaGemini, nor is Google called Alpha. More than that Alphabet is never shortened Alpha.
jcattle | 9 hours ago
Let's list them!
AlphaZero, AlphaGo, AlphaFold, AlphaGenome, AlphaEarth, AlphaGeometry, AlphaProteo, AlphaTensor.
There's a reason why so many people in the comments here thought that this was a model from Google.
Edit: Just looked through Wikipedia. Here's a few more: AlphaProof, AlphaDev, AlphaEvolve, AlphaStar, AlphaChip
seanhunter | 5 hours ago
Alpha Centauri (oh no, that's the star system).
Alpha-gal syndrome (oh no, that's the allergy to red meat you can get from being bitten by a tick).
Alpha-bits (oh that's the breakfast cereal)
Alpha course (oh no that's the thing for if you're Anglican and still want to feel like you're in a cult)
Alphabetti Spaghetti (actually I think that was Google Deepmind probably)
....
jcattle | 3 hours ago
If there's 10 breakfast cereals from the same company all starting with Alpha, I would give you some credit. But like this I don't even know what your reply is trying to tell me.
I honestly feel like I am taking crazy pills here.
seanhunter | 3 hours ago
And to answer your question directly, DEC had a whole series of things called "Alpha". eg AlphaStation family of workstations, the Alpha AXP series of processors etc (alpha 21064, alpha 21164, alpha 21264, alpha 21364).
williamdclt | 9 hours ago
jcattle | 9 hours ago
Let's list them!
AlphaZero, AlphaGo, AlphaFold, AlphaGenome, AlphaEarth, AlphaGeometry, AlphaProteo, AlphaTensor.
There are some revolutionary models in there. Especially AlphaZero and AlphaFold. AlphaFold even won a nobel price in 2024.
hansmoravec | 17 hours ago
romesc | 17 hours ago
Ensuring the assets are "CAD-level accurate" has a huge impact on making this work in practice. So much so, that our collaborators Yash Narang [1] and Iretiayo Akinola [2] have spent a number of years on just this - we are standing on the shoulders of giants!
Additionally, understanding the hardware well (including things like motor and controller dynamics) is non-negotiable. We learned the hard way that there is, in fact, no free lunch. You really have to get the deep understanding of physics and engineering right.
[1] https://scholar.google.com/citations?user=M3NuG7AAAAAJ&hl=en [2] https://scholar.google.com/citations?user=e1zesfMAAAAJ&hl=en
BoppreH | 21 hours ago
In case somebody got the same question as me: the robot "sees" point clouds as input, so basically the full simulation state except for velocities (correct me if I'm wrong). As put in the limitations section, it "[does] not establish robustness to perception error, latency, or contact-model mismatch".
squanch | 21 hours ago