Patrick

Computer Enjoyer

[JOJI @sushitrash]

demons

[Vitaly Bulatov @vitl2907]

Training robots just got TikTok easy.

Ultimate Bots Studio lets you make dances, gestures, full motion flows in 3 easy steps:

1) Upload a video → create a motion
2) Stitch motions → build a dance
3) Export to robot and share with community

Try it -> link in the comments

[UFB - Ultimate Fighting Bots @UFBots]

How do we train robots today? What’s changed so that anyone can teach them new moves?

This is what @DDobas broke down on the @TEDTalks stage today.

We’re moving from complex robotics pipelines to something much more intuitive: show the robot a move, and it learns.

The full talk drops in the coming weeks. For now, here are a few backstage moments from the UFB team in Vancouver, having way too much fun with it.

Thanks @bilawalsidhu for giving us the stage!

[annie @_annieversary]

what the Fuck https://t.co/jSWRRiEBQ5

[UFB - Ultimate Fighting Bots @UFBots]

New League. New Website.
Your favorite world-famous humanoid fights are entering the next phase.

We’re launching the next generation of humanoid sports - the UFB Fighting League.

Now’s your chance to get involved.
Visit https://t.co/L3WpmMQozJ and join as a Pilot, Ghost, or Team Manager.

[Vitaly Bulatov @vitl2907]

Yesterday we held the UFB Ghost Trials at @CalHacks Hack Nights — hands-on trials where students learned to retarget motions from videos, train motion imitation policies (BeyondMimic) in mjlab, then competed in sim + real robot deployments!

This is about making advanced humanoid training accessible to students everywhere — not just labs. Top teams impressed, making progress towards permanent placement in the league. Congrats to team Molt, Justin and Alice, for taking the win 🚀🤖

Shoutout to @kevin_zakka (creator of mjlab — the backbone of our sim training) and massive props to @UFBots tech team: @emerson @_patrickrose and @DDobas for powering this and pushing the frontier forward!

Video of sim showdowns, real deployments, and winners below 👇

[David Dobáš @DDobas]

Happy to have led the workshops at UFB Ghost Trials — covering video motion retargeting, training BeyondMimic in mjlab, latest papers like Sonic/BFM-Zero, plus team support, sim evals & real-robot deployments.

Thrilled seeing more students get hands-on with humanoid training! 🚀🤖

[David Dobáš @DDobas]

Last week, we ran 8 humanoid robots live at DJ Wukong’s Chinese New Year show.

5 on stage. 3 with VIPs.

I ran the full pipeline, from mocap to sim to real deployment.

Here’s what worked (and what didn’t).

[Patrick @_patrickrose]

@sreak1089 @pwlot @Sentdex [2/2]

Locomotion and proprioceptive coordination is one case where I suspect there is a significant contribution from evolutionary biases, with additional learning over the lifetime of an individual; see “motor babbling” in pre-term infants.

[Patrick @_patrickrose]

@sreak1089 @pwlot @Sentdex [1/2]

A guess: morphological and kinematic constraints on observations and actions provide a foundation. Broad biases in wiring preference (among many other factors) steer development towards certain structural outcomes, promoting useful perceptual and behavioral dynamics.

[UFB - Ultimate Fighting Bots @UFBots]

UFB Laboratory https://t.co/YN1qvr78Gs

[UFB - Ultimate Fighting Bots @UFBots]

wyd @KingJames?

[François Chollet @fchollet]

When you see the solution to AGI you will find that it was in fact so straightforward as to be obvious, and that it could have been developed decades ago

[Patrick @_patrickrose]

@alessiograncini @CalHacks True!

[Patrick @_patrickrose]

@StretchGoat_ https://t.co/GiFSjd3uhL

[Davide Scaramuzza @davsca1]

Impressive live demonstration by @UnitreeRobotics #G1 at the #IROS2025 conference! It really works!

[UFB - Ultimate Fighting Bots @UFBots]

the thrill of gladiator fights without a single human harmed in the process

[Taelin @VictorTaelin]

(Low quality opinion post / feel free to skip)

Now that AGI isn't cool anymore, I'd like to register the opposing position.

- AGI is coming in 2026, more likely than not

- LLMs are big memorization/interpolation machines, incapable of doing scientific discoveries and working on OOD concepts efficiently. They're not sufficient for AGI. My prediction stands regardless.

- Something akin to GPT-6, while not AGI, will automate human R&D to such extent AGI would quickly follow. Precisely, AGI will happen in, at most, 6 months after the public launch of a model as capable as we'd expect GPT-6 to be.

- Not being able to use current AI to speed up any coding work, no matter how OOD it is, is skill issue (no shots fired)

- Multiple paths are converging to AGI, quickly, and the only ones who do not see this are these focusing on LLMs specifically, which are, in fact, NOT converging to AGI. Focus on "which capabilities computers are unlocking" and "how much this is augmenting our own productivity", and the relevant feedback loop becomes much clearer.