Michelle Sun
@michellelsun
Physical AI value chain, Robotics, Industrial Automation
799 Following 11K Followers
.@Meta's 'one more thing' last night was the Muse Charm:
- a palm-sized AI holdable with a 2-inch OLED display, front and rear cameras, a built-in 5G modem.
- no WiFi needed, no phone tether
- push to talk using finger print
What I find most interesting is the use of the cameras. The cameras are more for Muse's continuous visual context for analysis and action, photo/video capture is secondary.
So, cameras are acting as the "eyes" for Muse first, and cameras for the user second.
That's a meaningful packaging decision for embodied perception. The Charm turns @muse into the always available agent, with eyes, ears and a 5G backhaul.
Show more
Amazing write up by David
My favorite was the anecdote 'Alibaba does a “worst product” award every year at all hands.'
It is such a great illustration of how American and Chinese education and culture differ
Show more
.@SkildAI ‘s S1 practiced soccer against increasingly capable versions of itself in simulation. The robot can now play against a human.
It’s an exciting glimpse of how robots might build on their existing skills
Congrats @deepakpathak and Skild team on another milestone !
Show more
We trained a robot to play football.
How? Self-play for 140 years in a virtual World Cup.
Meet the #Messinator#:
General world models could overtake specialized vision-language-action models around 2027.
@1x_tech founder @BerntBornich made this specific prediction in his interview with @ti_morse
Key highlights:
1. Data diversity is the constraint.
More recordings of the same tasks add less than exposure to new environments, objects, and ways tasks fail. In Bernt’s words, robotics is “diversity bound.”
2. Internet video could do most of the training.
Bernt expects it to provide roughly 99% of 1X’s training data. First-person human video, simulation, teleoperation, and robot experience would supply other kinds of examples.
3. The model should eventually collect its own training data. Bernt wants deployed robots to attempt unfamiliar tasks, record successes and failures, and improve through retraining without a human operating each attempt.
4. That gives the 50,000 target a second purpose.
1X aims to manufacture and ship 50,000 NEOs in 2027. A fleet that size could expose the model to far more variation, provided the robots meet the safety and quality bar for deployment.
Interview:
Show more
can someone older than 22 attend this? asking for a friend
@eriktorenberg @gaganbiyani
How @UnitreeRobotics newest Dex5-S compares with other Dexterous hands in the market
- Pricing
- Detailed specs
22-motor, 22-DOF hand at $6500
How does @UnitreeRobotics's Dex5-S hand compare with the @wuji_global Hand 2 and @SharpaRobotics SharpaWave?
Unitree vs. Wuji vs. Sharpa
- All 3 hands put one motor on each degree of freedom: 22, 20, 22
- All 3 hands are backdrivable
- Prices: $6500, $19500, $50,000
- Price per motor: $295, $975, $2270
Which one would you buy?
Show more
Unitree Introducing: Unitree Dex5-S Dexterous Hand 22 Degrees of Freedom 1:1 Real-Hand Size👋
Precision biomimetic dexterous hand, price from $6.5K (Tax and Shipping cost excluded), with all 22 joints supporting smooth backdrivability, and each joint equipped with limit impact torque protection.
Show more
Scott Walter, Robotics Research Diligence Director at RoboStrategy ($BOT), wants a humanoid decathlon.
“I'm throwing down the gauntlet for the humanoid decathlon challenge.”
One robot. The same 10 human events. The same rules.
That means more than a sprint demo:
› Hurdles and turns
› A pole vault that tests upper-body and grip strength
› A baton exchange that demands manipulation
› A long-jump sand pit, because sand in the joints is part of the test
A specialized robot can win one optimized event. A decathlon asks which joints, control policies, and hands survive across a body of work.
@GoingBallistic5 @RoboStrategy
Show more
While we do a quick survey of the main hand designs on the market, it is primarily used to contrast three designs less talked about. I appreciate Michelle coming up with the idea to discuss Allonic, Daxo and Tacta.
Show more
Inside RoboStrategy $BOT's robotics research with Scott Walter, PhD @GoingBallistic5
Tesla, Figure and the robot hand fight.
“If they exist, where are they?”
Scott breaks down Tesla Optimus (@Tesla_Optimus), Figure (@Figure_robot), 1X (@1X_tech), Wuji (@Wuji_global) and Sharpa Wave (@SharpaRobotics), then goes to the design extremes:
a braided hand, up to 120 tendons, and a hydraulic manipulation system built outside the humanoid.
“They both have their pros and cons. They both have the big engineering challenges in what they're trying to do.”
We cover:
› Tesla and Figure's opposing paths through tendons
› Why a wrist makes cable routing much harder
› @AllonicRobotics, @DaxoRobotics and @TactaSystems at the design extremes
› A decathlon for humanoids
› Why robot runners may stop looking human
› General-purpose bodies and specialized workers
TIMESTAMPS
00:00 The labor prize and the 5-year forecast
00:24 The Fermi paradox of robotic hands
01:18 Tesla, Figure and the tendon debate
05:10 Bowden tubes and wrist routing
11:15 Joints, actuators and real control
14:38 Why the pinky matters
17:23 Humanoid Games as engineering experiments
21:57 The humanoid decathlon challenge
26:28 Tesla's abandoned hand iterations
27:55 Wuji and Sharpa Wave
38:29 Three extreme hand designs
38:53 Allonic's braided structure
51:15 Daxo and up to 120 tendons
01:02:36 Tacta's worker data
01:05:18 A hand system outside the humanoid
01:13:12 Data-center cable insertion
01:16:22 General bodies, specialized work
01:19:18 Why this 5-year forecast may be different
Show more
Inside RoboStrategy $BOT's robotics research with Scott Walter, PhD @GoingBallistic5
Tesla, Figure and the robot hand fight.
“If they exist, where are they?”
Scott breaks down Tesla Optimus (@Tesla_Optimus), Figure (@Figure_robot), 1X (@1X_tech), Wuji (@Wuji_global) and Sharpa Wave (@SharpaRobotics), then goes to the design extremes:
a braided hand, up to 120 tendons, and a hydraulic manipulation system built outside the humanoid.
“They both have their pros and cons. They both have the big engineering challenges in what they're trying to do.”
We cover:
› Tesla and Figure's opposing paths through tendons
› Why a wrist makes cable routing much harder
› @AllonicRobotics, @DaxoRobotics and @TactaSystems at the design extremes
› A decathlon for humanoids
› Why robot runners may stop looking human
› General-purpose bodies and specialized workers
TIMESTAMPS
00:00 The labor prize and the 5-year forecast
00:24 The Fermi paradox of robotic hands
01:18 Tesla, Figure and the tendon debate
05:10 Bowden tubes and wrist routing
11:15 Joints, actuators and real control
14:38 Why the pinky matters
17:23 Humanoid Games as engineering experiments
21:57 The humanoid decathlon challenge
26:28 Tesla's abandoned hand iterations
27:55 Wuji and Sharpa Wave
38:29 Three extreme hand designs
38:53 Allonic's braided structure
51:15 Daxo and up to 120 tendons
01:02:36 Tacta's worker data
01:05:18 A hand system outside the humanoid
01:13:12 Data-center cable insertion
01:16:22 General bodies, specialized work
01:19:18 Why this 5-year forecast may be different
Show more
Foundation world model that can control robots, drive cars, and play video games.
I tried Odyssey's earlier model a few months ago and was impressed how well the video generates from a prompt. This is an exciting milestone.
Congratulations @olivercameron @jeffrey_hawke @odysseyml team!
Show more
Today we’re unveiling Odyssey-3, a big step forward for foundation world models.
It can control robots, power humanoids, drive cars (on the roads of India!), train AIs, pilot drones, and even play video games.
We can’t wait to see what intelligent systems it enables.
Show more
Is Astra the ChatGPT moment for Robotics? A few insights:
1) Strong in-context learning
In 3 tries, it went from learning the controls and understanding the environment to accomplishing a task.
The model reasons through its mistakes; notice the change in grip angle and carry height. It's slow (more on this later) but it's clearly understanding the controls and the space in which it's interacting. 🧵(1/5)
Show more
A weekend exercise for learning robotics: pick one component and explain what goes into it, what comes out, and what changes if you remove it.
Then follow the connection to the next component.
How far can you get before you need to look something up?
Show more
Inside Prensilia w/ Managing Director Francesco Clemente
Why fingers fail before motors or gears in a robotic hand.
"Fingers are the parts that break the most, because you have impacts with objects."
Prensilia is a Pisa, Italy hand-design firm founded in 2009. Its Mia hand carries three motors and has survived
300,000 grasp cycles at full grip force in company testing.
"So we really try to have high robustness with our devices."
"We are talking about thousands of cycles, not really millions of cycles, with the tendons."
Show more
There's something calming about watching a robot paint with a brush. @cdngdev
Curious what kind of feedback helped it improve between attempts.
Would be fun to make it a series of paintings!
i gave astra a robot, a paint brush, and a camera then asked it to paint the golden gate bridge in real life!
it figured out how to control the robot, and progressively got better throughout its attempts. the timelapse is sick
Show more
A well articulated piece on how @SkildAI approaches deployment:
1) rewarding deployment rather than demo to build the right company culture
2) think of S1 as the high school brain and each deployment helps it build the specialized knowledge
Congrats @deepakpathak and S1 team on this milestone!
Show more
Great job @earthtojake - mesmerizing to look at
Inside Unitree's 3-hour IPO roadshow:
Wang Xingxing and Unitree's management team answered 363 investor questions. Here are 5 highlights:
1. Industrial work was 9% of humanoid revenue, and more than 70% came from universities and research institutes.
2. 1Q2026 profit fell 52.55%, and 3 of the 5 reasons management gave were about weakening demand.
3. Unitree's actuator design appears to skip harmonic reducers, one of the most expensive parts in most humanoid cost models.
Asked what share it makes in-house, Wang Xingxing answered without using the word harmonic once.
4. Unitree has released five AI models in eleven months. One deployment has gone live so far, inside its own factory.
DeepSeek bought Rmb141m in the offering, and management did not disclose details on what the two are building together.
5. Unitree’s 6 robot models (quadrupeds and humanoids) hold FCC certification and can still be sold in the US.
Full transcript (>50,000 words) in today's Core Matter.
Show more
