Hey friend,
Welcome to yet another Friday Robotics issue of withAgents - hope you had a great week. (And you’re right, no, I’ve never actually done a Friday Robotics issue before. It just sounded more official to start that way.)
So why robotics?
Because Physical Intelligence.
This is a fast-growing hotbed of innovation. It’s debatably the most important, world-changing space for applied AI in the coming century and will solve some amazing problems.
But the news here is also often drowned out by debates over chat-model performance and white-collar job security. I think it deserves its own corner in our weekly reporting, at minimum.
Let me know what you think, bleep bloop,
Clay
The big robotic story of the week
“Robots for babies and robots for old folks”
First things first.
To answer my headline question “Wait, we really have Surrogacy Robots now?” — Nah. We don’t.
I look at LOTS of sources to build these reports. There was one story that I saw over and over again this week. A juicy one. About a Chinese company’s announcement, at the World Robot Expo in Beijing earlier this month. They’d made (and lab-proven!) a surrogacy robot. Fully capable of gestating a baby, with synthetic amniotic fluid, tubes and frequent parent/fetus interaction to simulate natural gestation.
Obviously this seems off, but with Nurse.org, Newsweek, The Jerusalem Post, Yahoo, The Economic Times and tens more talking about it… could it be?
Thankfully, Snopes did the dirty work to debunk this one. Just good old fashion Sinophobia.
But China did create a lot of true news in robotics this week. Like their expansion of robot fleets for US nursing home deployment.
“Hi Betty, I’ve brought your morning pills.”
Source: Joan Cros/NurPhoto via Getty Images via Forbes
130 of these small care-bots already generated $1.8M for Robocore through deployment in New York state nursing homes. They visit patients, facilitate doctor/patient televisits and can run basic wellness diagnostics.
Now, Foxconn, the world’s largest electronics manufacturer, has invested another $10M to drive further expansion and R&D for Robocore.
China also ran the first ever Robot Olympics this week, with Tesla’s biggest robot competitor competing.
More on the great robot games in our snapshots below.
Expect to see Chinese innovation frequently discussed in our Friday Robotics issues, they appear to have a leg up in this space. The National Security Journal published a call for a “China Tech Power Report” by Congress, and Kharon did an in-depth report on academic drone research and its ties to military programs in the region.
The US must put better focus here too.
Interesting human/AI news
Past week
“World” Humanoid Robot Games showcase China’s AI vision
Strongly considering spending my weekend with game-day snacks and this on the teli. It was Beijing’s inaugural Humanoid Robot Games where tele‑operated and autonomous robots duked it out. There was robot sprints, boxing and more. Unitree’s H1 set a 100‑m record and close to 200 university teams and 88 companies participated. Loved the top Youtube comment: “10 yrs from now will be robots watching human fight in a cage” - @jwuhome
Human‑in‑the‑loop reinforcement learning trains robots in hours
UC Berkeley’s “HiL‑SERL” (Elon’s next kid’s name) method mixes human-driven demonstrations and human interventions. Robots learned tasks like flipping an egg, moving Jenga blocks and assembling motherboards quickly, achieving 100 % success within 1‑2 hours.Tech imitates nature, researchers take inspo from water bug feet
Georgia Tech researchers mimicked the ribbon‑like fans on water bug feet, to create a 1‑mg fan that opens and closes via surface tension, letting an insect‑sized robot accelerate, brake and turn on water“Large Behavioral Model” (LBM) test @ Boston Dynamics. Very cool.
“Hil-SERL” sounds neat, but in terms of news in robotics training, we like this news more — if only for the cool video and big promise. The basic idea here is that with an AI model trained on general robotic motions, the robot can rapidly learn necessary motions for its body and scale that flexibility without direct training. Check it.
Using AR to give robots more emotive, reactive expressions
This one caught our eye. The idea is, take a basic robot (in this case a simple robotic pet - like a robotic dog) and mix that with AR-goggles that place a face on top of that robot. The face interacts, reacts, interprets mood, has motion when speaking, etc. See a video in the link.Wearable robots for limb impairment, trained on individual motion
Harvard showed off improvements in their wearable robot, which is worn by victims of strokes or ALS, allowing them to move their limbs more naturally and with less effort. The update trains the robot on the individual’s own motions, making its performance more personal to the wearer.
I'd love to hear from you!
Thanks as always, human.
