mvandepanne.bsky.social - Profile | ThreadSky | a Reddit-style client for Bluesky

Impressive learned stand-up policies for Unitree G1. Interesting use of an assistive-force curriculum, to allow for more effective discovery of solution modes. www.youtube.com/watch?v=Yruh... Project page: taohuang13.github.io/humanoid-sta...

submitted 10 days ago • 0 comments

I really enjoyed this thought provoking talk… thanks @natolambert.bsky.social !

submitted 12 days ago • 0 comments

And that's right -- the annualized mean growth rate for CO2 is now at a record high 3.55 ppm as of January, 2025. Atmospheric CO2 is growing at the fastest rate in recorded history. This is not going to end well for global industrial civilization, children and other living things.

submitted 19 days ago • 22 comments

Manually fixing broken multiplayer interaction animations is challenging. One of the main benefits of physics-based motion cleanup is that it is straightforward to extend to multiplayer—env code requires minimal modification, training code is unchanged, but training does take longer.

submitted 20 days ago • 2 comments

We've built a simulated driving agent that we trained on 1.6 billion km of driving with no human data. It is SOTA on every planning benchmark we tried. In self-play, it goes 20 years between collisions.

submitted 20 days ago • 22 comments

No they don’t. Canada and Mexico want none of this. They want the president of the United States to respect and abide by the trade treaty he negotiated with them - and not make up bogus pretexts for plunging us all into a gratuitous no win economic debacle.

submitted 24 days ago • 428 comments

Here's the speech if you haven't heard it. 13 minutes of your time. Share it with your American friends and neighbours, because they won't see it otherwise. Circulate it in Mexico and Denmark, France and the UK, Panama and Germany, Poland and Estonia, and especially -- especially -- Ukraine.

submitted 25 days ago • 195 comments

The billionaire tech coup of social media is an SOS, so here is FOF: Free Our Feeds! We are not powerless. Please join me in following and funding them here: 💙 bsky.app/profile/free... 💙 💎 freeourfeeds.com 💎

submitted 44 days ago • 2 comments

The hard truth about the consequences of the #ClimateCrisis — wildfires, floods, storms, droughts, heat waves and more — won’t come from climate activists who are too easily ignored or even vilified. It will come from central banks. And insurance companies. And every rational evidence-based sector.

submitted 49 days ago • 41 comments

For Maths fans, 2025 is a square. 45² = 45 x 45 = 2025 Also, 9² x 5² = 2025 40² + 20² + 5² = 2025 My favourite? 1³+2³+3³+4³+5³+6³+7³+8³+9³ = 2025 #Mathematics #teaching #education

submitted 56 days ago • 955 comments

Every Jan 1 I post a scorecard on predictions I made, with dates, on Jan 1, 2018 on cars (self-driving), robots, AI, & ML, and on human spaceflight. Besides telling which turned out right and which wrong in the last year I also talk a lot of smack about these topics. rodneybrooks.com/predictions-...

submitted 57 days ago • 8 comments

I have created a list of open research problem from my work and from writing our textbook. Something to work on for the new year! Comments please! #relationallearning #causality #AI #aifca #starAI www.cs.ubc.ca/~poole/resea...

submitted 57 days ago • 2 comments

It's been a while since I did a pure Machine Learning article on my blog, but today I have a post for you about adding noise to Neural Networks, and some fun experiments with Flow-Matching: theorangeduck.com/page/noise-n...

submitted 59 days ago • 1 comment

What happens when techno-optimism meets energy system realism? My latest for @BloombergNEF is a long read on AI, called The Power and the Glory. So that's Boxing Day sorted! about.bnef.com/blog/liebrei...

submitted 63 days ago • 22 comments

Due to this idiotic statement, by a person who has proven his inability to distinguish truth from online propaganda, I just donated to Wikipedia. Please do the same.

submitted 63 days ago • 101 comments

The story of the Genesis library is about how the relentless quest for hype by senior people and institutions ends up completely overshadowing what is a great new simulation tool, making us all more cynical in the process. arstechnica.com/information-...

submitted 67 days ago • 4 comments

Yesterday the hyped Genesis simulator released. But it's up to 10x slower than existing GPU sims, not 10-80x faster or 430,000x faster than realtime since they benchmark mostly static environments blog post with corrected open source benchmarks & details: stoneztao.substack.com/p/the-new-hy...

submitted 68 days ago • 4 comments

"Sora is a data-driven physics engine." x.com/chrisoffner3...

submitted 78 days ago • 12 comments

If you're at NeurIPS, RLC is hosting an RL event from 8 till late at The Pearl on Dec. 11th. Join us, meet all the RL researchers, and spread the word!

submitted 78 days ago • 2 comments

RLC will be held at the Univ. of Alberta, Edmonton, in 2025. I'm happy to say that we now have the conference's website out: rl-conference.cc/index.html Looking forward to seeing you all there! @rl-conference.bsky.social #reinforcementlearning

submitted 96 days ago • 2 comments

Great to see progress continue on interactive diagrams -- paper from UIST 2024: Creating Interactive and Embedded Physics Simulations from Static Textbook Diagrams ryosuzuki.org/augmented-ph... www.youtube.com/watch?v=HUgY...

submitted 92 days ago • 0 comments

New here? Interested in AI/ML? Check out these great starter packs! AI: go.bsky.app/SipA7it RL: go.bsky.app/3WPHcHg Women in AI: go.bsky.app/LaGDpqg NLP: go.bsky.app/SngwGeS AI and news: go.bsky.app/5sFqVNS You can also search all starter packs here: blueskydirectory.com/starter-pack...

submitted 110 days ago • 70 comments

I love that traditional CGI obeys the laws of physics but not the laws of the statistical structure of naturalistic images and generative video obeys the latter but not the former, leading to two equally but differently uncanny modes of video generation.

submitted 99 days ago • 2 comments

It is so incredibly nice to have a social platform that values links again - a breath of fresh air in the poisoned ecosystem of the creator economy

submitted 100 days ago • 110 comments

Nvidia's LLaMA-Mesh! Nvidia fine-tuned LLaMA on 3D Mesh data, enabling LLMs to natively generate 3D meshes via chatting while retaining their original language capabilities.

submitted 100 days ago • 2 comments

A short thread on BlueSky, translated from my dear friend @beaudoin.social [start] 1/ “It may be decentralized, but it’s the same! » Why Bluesky has a chance of becoming truly different.* * Informed of nearly a thousand hours spent tinkering with the code of the beast.

submitted 103 days ago • 1 comment

Recent paper of interest -- InterControl: text-conditioned interaction of multiple characters using diffusion models. Is trained only on single-character motions, with the help of IK-guidance. NeurIPS 2024 poster. openreview.net/pdf?id=AH1mF...

submitted 101 days ago • 0 comments

A physical reasoning agent, trained on millions of problems, that you can test out by coming up with physics puzzles and seeing if it can solve them. It's fun to see if you can stump it: kinetix-env.github.io

submitted 102 days ago • 5 comments

Little known research hack that should be appreciated more!

submitted 101 days ago • 0 comments

Clearly though, sim to real is the way of the future. Even these guys are doing their locomotion via sim to real! Unitree is doing it! This is the economical, scalable way to train robots

submitted 105 days ago • 2 comments

If you're an RL researcher or RL adjacent, pipe up to make sure I've added you here! go.bsky.app/3WPHcHg

submitted 109 days ago • 51 comments

Any suggestions for the best ways to bookmark or otherwise log particular posts for future reference?

submitted 110 days ago • 0 comments

Chips specialized for Transformers www.etched.com/announcing-e...

submitted 111 days ago • 0 comments

Today's retro video -- a voice driven animation system that we had built back in 2006. NLP has advanced so much since then. youtu.be/gY50LuKQLXQ

submitted 111 days ago • 0 comments

Control physics-based characters at interactive rates to climb over tables and more. No learning! Key ideas: (a) partwise MPC: switch between full-body and factored body-part MPC as convenient; (b) strategy guidance via sparse contact keyframes. SCA 2024 www.cs.ubc.ca/~van/papers/...

submitted 111 days ago • 0 comments

Full-Order Sampling-Based MPC for Torque-Level Locomotion Control via Diffusion-Style Annealing DIAL-MPC (from CMU) lecar-lab.github.io/dial-mpc/

submitted 111 days ago • 0 comments

Impressive learned stand-up policies for Unitree G1. Interesting use of an assistive-force curriculum, to allow for more effective discovery of solution modes. www.youtube.com/watch?v=Yruh... Project page: taohuang13.github.io/humanoid-sta...

I really enjoyed this thought provoking talk… thanks @natolambert.bsky.social !

And that's right -- the annualized mean growth rate for CO2 is now at a record high 3.55 ppm as of January, 2025. Atmospheric CO2 is growing at the fastest rate in recorded history. This is not going to end well for global industrial civilization, children and other living things.

Manually fixing broken multiplayer interaction animations is challenging. One of the main benefits of physics-based motion cleanup is that it is straightforward to extend to multiplayer—env code requires minimal modification, training code is unchanged, but training does take longer.

We've built a simulated driving agent that we trained on 1.6 billion km of driving with no human data. It is SOTA on every planning benchmark we tried. In self-play, it goes 20 years between collisions.

No they don’t. Canada and Mexico want none of this. They want the president of the United States to respect and abide by the trade treaty he negotiated with them - and not make up bogus pretexts for plunging us all into a gratuitous no win economic debacle.

Here's the speech if you haven't heard it. 13 minutes of your time. Share it with your American friends and neighbours, because they won't see it otherwise. Circulate it in Mexico and Denmark, France and the UK, Panama and Germany, Poland and Estonia, and especially -- especially -- Ukraine.

The billionaire tech coup of social media is an SOS, so here is FOF: Free Our Feeds! We are not powerless. Please join me in following and funding them here: 💙 bsky.app/profile/free... 💙 💎 freeourfeeds.com 💎

For Maths fans, 2025 is a square. 45² = 45 x 45 = 2025 Also, 9² x 5² = 2025 40² + 20² + 5² = 2025 My favourite? 1³+2³+3³+4³+5³+6³+7³+8³+9³ = 2025 #Mathematics #teaching #education

Every Jan 1 I post a scorecard on predictions I made, with dates, on Jan 1, 2018 on cars (self-driving), robots, AI, & ML, and on human spaceflight. Besides telling which turned out right and which wrong in the last year I also talk a lot of smack about these topics. rodneybrooks.com/predictions-...

I have created a list of open research problem from my work and from writing our textbook. Something to work on for the new year! Comments please! #relationallearning #causality #AI #aifca #starAI www.cs.ubc.ca/~poole/resea...

It's been a while since I did a pure Machine Learning article on my blog, but today I have a post for you about adding noise to Neural Networks, and some fun experiments with Flow-Matching: theorangeduck.com/page/noise-n...

What happens when techno-optimism meets energy system realism? My latest for @BloombergNEF is a long read on AI, called The Power and the Glory. So that's Boxing Day sorted! about.bnef.com/blog/liebrei...

Due to this idiotic statement, by a person who has proven his inability to distinguish truth from online propaganda, I just donated to Wikipedia. Please do the same.

The story of the Genesis library is about how the relentless quest for hype by senior people and institutions ends up completely overshadowing what is a great new simulation tool, making us all more cynical in the process. arstechnica.com/information-...

Yesterday the hyped Genesis simulator released. But it's up to 10x slower than existing GPU sims, not 10-80x faster or 430,000x faster than realtime since they benchmark mostly static environments blog post with corrected open source benchmarks & details: stoneztao.substack.com/p/the-new-hy...

"Sora is a data-driven physics engine." x.com/chrisoffner3...

If you're at NeurIPS, RLC is hosting an RL event from 8 till late at The Pearl on Dec. 11th. Join us, meet all the RL researchers, and spread the word!

RLC will be held at the Univ. of Alberta, Edmonton, in 2025. I'm happy to say that we now have the conference's website out: rl-conference.cc/index.html Looking forward to seeing you all there! @rl-conference.bsky.social #reinforcementlearning

Great to see progress continue on interactive diagrams -- paper from UIST 2024: Creating Interactive and Embedded Physics Simulations from Static Textbook Diagrams ryosuzuki.org/augmented-ph... www.youtube.com/watch?v=HUgY...

New here? Interested in AI/ML? Check out these great starter packs! AI: go.bsky.app/SipA7it RL: go.bsky.app/3WPHcHg Women in AI: go.bsky.app/LaGDpqg NLP: go.bsky.app/SngwGeS AI and news: go.bsky.app/5sFqVNS You can also search all starter packs here: blueskydirectory.com/starter-pack...

I love that traditional CGI obeys the laws of physics but not the laws of the statistical structure of naturalistic images and generative video obeys the latter but not the former, leading to two equally but differently uncanny modes of video generation.

It is so incredibly nice to have a social platform that values links again - a breath of fresh air in the poisoned ecosystem of the creator economy

Nvidia's LLaMA-Mesh! Nvidia fine-tuned LLaMA on 3D Mesh data, enabling LLMs to natively generate 3D meshes via chatting while retaining their original language capabilities.

A short thread on BlueSky, translated from my dear friend @beaudoin.social [start] 1/ “It may be decentralized, but it’s the same! » Why Bluesky has a chance of becoming truly different.* * Informed of nearly a thousand hours spent tinkering with the code of the beast.

Recent paper of interest -- InterControl: text-conditioned interaction of multiple characters using diffusion models. Is trained only on single-character motions, with the help of IK-guidance. NeurIPS 2024 poster. openreview.net/pdf?id=AH1mF...

A physical reasoning agent, trained on millions of problems, that you can test out by coming up with physics puzzles and seeing if it can solve them. It's fun to see if you can stump it: kinetix-env.github.io

Little known research hack that should be appreciated more!

Clearly though, sim to real is the way of the future. Even these guys are doing their locomotion via sim to real! Unitree is doing it! This is the economical, scalable way to train robots

If you're an RL researcher or RL adjacent, pipe up to make sure I've added you here! go.bsky.app/3WPHcHg

Any suggestions for the best ways to bookmark or otherwise log particular posts for future reference?

Chips specialized for Transformers www.etched.com/announcing-e...

Today's retro video -- a voice driven animation system that we had built back in 2006. NLP has advanced so much since then. youtu.be/gY50LuKQLXQ

Control physics-based characters at interactive rates to climb over tables and more. No learning! Key ideas: (a) partwise MPC: switch between full-body and factored body-part MPC as convenient; (b) strategy guidance via sparse contact keyframes. SCA 2024 www.cs.ubc.ca/~van/papers/...

Full-Order Sampling-Based MPC for Torque-Level Locomotion Control via Diffusion-Style Annealing DIAL-MPC (from CMU) lecar-lab.github.io/dial-mpc/

A highly recommended read, for anyone working on animating human motion. theorangeduck.com/page/animati...

The robotics renaissance continues! www.youtube.com/watch?v=9Lyp...

Great set of invited speakers! 2024.ieee-humanoids.org/plenaries/

Looking forward to creating critical mass (and critical thinking) here.