Christian Hubicki

How do their cybersecurity vulnerabilities factor into this? We firewall our Unitree.

And by "some" they mean people who have no clue about how AI works.

Makes me curious what the up and down times are.

Yeah, the reliability is always what I’m looking for.

Killer Robots, AI Psychosis and Nuclear War: The Pentagon’s Biggest AI Fears A Q&A with a former Pentagon insider on the AI debates that could shape the future of national security.

The interview’s vague, but the Pentagon is considering Generative AI for decision support — a terrible idea.

If LLMs are ever helpful it’s in *low*-stakes suggestions, not life-or-death choices. But for now, industry guardrails hinder LLM use in lethal scenarios.

www.politico.com/news/magazin...

*tele-operation

In this case, teleportation would actually be more impressive than a pre-canned trained routine.

Aaaaannd it’s already ruined.

It’s true the controller is a neural network, but the routine is pre-choreographed. It’s not responding to the human and making decisions how to “fight,” (it’s probably not even using vision) which is of course what the typical person would assume by saying it’s “AI.”

Basically, good job. But I’m also judging your robot on the company claim that it will be a multi-TRILLION dollar product that we’ll send to Mars soon.

But it feels to me a fully separate endeavor from making humanoids useful in the near term. This control method is kinda hard to marry with the dexterous task controllers (stable diffusion) humanoids are trying to use.

It’s more saying, “we can do king fu too, Unitree”.

I’ll highlight the best part—the balance.

They train it in a computer simulation, pushing it at random times during while attempting to execute the performance. That’s why it can take a shove, and can shimmy on one foot to rebalance.

While not fully new, it’s uncommon for a full-size humanoid.

If it’s like other companies, they’re doing motion retargeting using deep reinforcement learning (DRL). That’s different than their typical balancing which is more traditional.

It’s cleaner than their prior DRL results👇, but likely because it’s designed to imitate the symmetry of human performance.

Good martial arts routine from Tesla.

It’s trained to do a pre-choreographed routine and the human is playing along.

I’ll wonk out more in the replies, but in short, their balance control is getting much better (other companies have too) but this is mostly a party trick.

Okay, the AI actor thing is ridiculous. It’s a fully synthetic media push. I regret engaging with the articles even to poke fun at them.

So glad that professors can get family leave. With time off, I can finally give the care, attention, and patience that my grad students need.

Jane Goodall, Eminent Primatologist Who Chronicled the Lives of Chimps, Dies at 91

Very sorry to hear of Jane Goodall’s passing. Her scientific contributions and her work as a conservationist are extraordinary.
#goodall #chimpanzee #conservation #primatology

🧪

Gift article

Elon Musk’s Optimus Robots Face Speedy Downfall On Mars, Scholar Says SpaceX’s Elon Musk says he will rocket Optimus bots on the first Starships sent to Mars, but a leading robotics scholar predicts their space odyssey will be short-lived.

The scholar is me. Humanoids are not ready for Mars.

www.forbes.com/sites/kevinh...

Wikipedia is great at setting standards for math articles. I can always expect a page that’s beautiful, thorough, and unhelpful.

Photo from a different era. My wife finds it so funny that it’s on our refrigerator.

A view from above of a large cyclonic storm that is mainly a grey colour, set against a swirling backdrop of more bluish smaller storms. At the 12 o'clock position in the large storm is a small point of green light, which is a lightning bolt. This scene is on Jupiter.

This is a lightning bolt in a storm.

On Jupiter.

Sign the petition to restore eligibility of 2nd year grad students to the NSF GRFP. They are being robbed of the opportunities to advance their innovative science. @chubicki.bsky.social

It’s needlessly unfair. They could have avoided a lot of heartache by just phasing it in over a year.

Done!

Thankfully there’s still time to reverse the decision. I hope they reconsider— it costs the program nothing to change it back for at least this year.

Poor students got the rug pulled from under them.

The National Science Foundation’s premiere funding award for grad students shrank their eligibility pool without warning.

1st year PhD students who waited until their 2nd to submit their once-ever application were just told they’re not eligible.

Working through the latest lengthy #robotics piece from Rodney Brooks and literally jump-scared when I saw my name a third of the way through.

Creator of AI Actress Tilly Norwood Responds to Backlash: “She Is Not a Replacement for a Human Being” Eline Van der Velden said AI characters "should be judged as part of their own genre [...] rather than compared directly with human actors," after news broke her studio is seeking representation for N...

The obvious countermove to commoditizing AI-generated actors is to train your own AI models to impersonate them.

Getting cease-and-desisted for appropriating the likeness of an AI model made by stealing thousands of likenesses would underscore the point.

www.hollywoodreporter.com/movies/movie...

Beautiful.

It's unclear if these security vulnerabilities are intentional or just mistakes.. but either isn't good.

And by everywhere, I mean these things are everywhere. Here's one strutting around West Hollywood in pride decor.

Karalyn Davis via Storyful.
www.youtube.com/shorts/OveZn...

This is the Unitree G1. You're seeing these robots all over because they're cheap (less than a car).

But techies keep finding security flaws: they secretly stream location data to China--and might be susceptible to remote takeover.

spectrum.ieee.org/unitree-robo... via @evanackerman.bsky.social