Pinned Tweet
So glad we got to visit so many amazing robotics companies together.
Hope you come back every year. This industry moves fast and never disappoints.
There are so many great robotics companies here, and everyone is welcome to come see them.
Hope you come back every year. This industry moves fast and never disappoints.
There are so many great robotics companies here, and everyone is welcome to come see them.
That's a wrap on an amazing 17-day journey across China.
I didn't do this alone. My friend @dolylupec came along for the whole ride. She's got a background in AI and robotics investing, and I'm grateful she chronicled the tours on her account and pitched in on my channel.
e most resourceful people I know. He understands the humanoid space well and helped us schedule most of the important visits, then came along too.
And a shout-out to @Robo_Tuo, who encouraged me to do this trip and supported us all along.
It was my first time in China, and the trip was packed so tight with tours, we didn’t get to do much touristy stuff. Can't complain. It was a hell of a ride.
4 cities on the mainland. 18 humanoid makers. Warm, welcoming people everywhere. We had the privilege of chatting with founders, senior engineers, researchers, and marketing and strategy leads, a lot of whom said yes on very short notice. Grateful for that.
The energy is extraordinary. It's the trifecta: engineering talent, a deep supply chain, and breakneck adoption of new tech. There's intense competition, but underneath it, a real spirit of sharing and collaboration.
I've got a series of tour videos lined up, dropping over the coming days.
And something tells me this won't be my last trip. Robotics and embodied AI are just getting started. The next ten years are going to reshape how we think about physical work.
We ended on the Great Wall, a 2,000 year old marvel of engineering, after two weeks staring at the next one.
The robots are coming. I just went to meet them first.
5
6
54
27,811
Another robot horse… this one’s built to carry your gear and run your errands.
PHYMI just unveiled Sagi, a robotic centaur with a humanoid upper body, four wheeled legs and a cargo box on its back. The box carries the load between stops, leaving its arms free to handle things.
t to work hauling camping equipment, fetching items around an office and helping a courier with deliveries.
PHYMI also describes LINK, a system for giving Sagi tasks and helping it when a decision needs your input. You can see what it sees, clarify what you want and let it continue the errand.
PHYMI just unveiled Sagi, a robotic centaur with a humanoid upper body, four wheeled legs and a cargo box on its back. The box carries the load between stops, leaving its arms free to handle things.
t to work hauling camping equipment, fetching items around an office and helping a courier with deliveries.
PHYMI also describes LINK, a system for giving Sagi tasks and helping it when a decision needs your input. You can see what it sees, clarify what you want and let it continue the errand.
3
7
48
1,458
And here’s the off-road version, for when nature calls and you can’t wait. 😂
This robotic toilet might be more useful than a robot that does your chores.
China’s Yueban built Xiaoban for people with limited mobility. Call it by voice or remote, and it comes to the bedside to help them use the toilet.
It plans its own route, detects obstacles and Once in position, it locks its wheels and deploys supports so users can lean on the armrests as they sit down or stand up.
It has a warm-water bidet and a warm-air dryer. Afterward, it heads to the bathroom, pumps waste into the existing toilet and rinses out its tank. No commode bucket for a caregiver to carry, empty or scrub.
Yueban says the camera stays off while someone is using Xiaoban. It only turns on to locate the bathroom toilet during emptying.
For older adults who struggle to get to the bathroom, especially in the middle of the night, this could be a huge relief.
5
3
22
1,268
While humanoid makers work to bring robots into our homes, Japan’s MW plans to sell homes with a housework robot hanging from the ceiling.
In MW’s design, MW bot travels along ceiling and wall rails, keeping its base off the floor. It uses two arms for chores, works alongside MW tucks away in a Nest when idle.
Before anyone moves in, MW can plan counter heights, laundry stations and travel routes around the robot’s reach. That can reduce the variation it has to handle.
MW demonstrated a prototype folding laundry in September, but it was teleoperated. The company is targeting an autonomous home demo in 2027 and sales of robot-equipped homes in 2028.
For new construction, this approach makes sense. The trade-off is the installation work and a working area defined by the rails.
In MW’s design, MW bot travels along ceiling and wall rails, keeping its base off the floor. It uses two arms for chores, works alongside MW tucks away in a Nest when idle.
Before anyone moves in, MW can plan counter heights, laundry stations and travel routes around the robot’s reach. That can reduce the variation it has to handle.
MW demonstrated a prototype folding laundry in September, but it was teleoperated. The company is targeting an autonomous home demo in 2027 and sales of robot-equipped homes in 2028.
For new construction, this approach makes sense. The trade-off is the installation work and a working area defined by the rails.
10
10
65
4,319
This robotic toilet might be more useful than a robot that does your chores.
China’s Yueban built Xiaoban for people with limited mobility. Call it by voice or remote, and it comes to the bedside to help them use the toilet.
It plans its own route, detects obstacles and Once in position, it locks its wheels and deploys supports so users can lean on the armrests as they sit down or stand up.
It has a warm-water bidet and a warm-air dryer. Afterward, it heads to the bathroom, pumps waste into the existing toilet and rinses out its tank. No commode bucket for a caregiver to carry, empty or scrub.
Yueban says the camera stays off while someone is using Xiaoban. It only turns on to locate the bathroom toilet during emptying.
For older adults who struggle to get to the bathroom, especially in the middle of the night, this could be a huge relief.
China’s Yueban built Xiaoban for people with limited mobility. Call it by voice or remote, and it comes to the bedside to help them use the toilet.
It plans its own route, detects obstacles and Once in position, it locks its wheels and deploys supports so users can lean on the armrests as they sit down or stand up.
It has a warm-water bidet and a warm-air dryer. Afterward, it heads to the bathroom, pumps waste into the existing toilet and rinses out its tank. No commode bucket for a caregiver to carry, empty or scrub.
Yueban says the camera stays off while someone is using Xiaoban. It only turns on to locate the bathroom toilet during emptying.
For older adults who struggle to get to the bathroom, especially in the middle of the night, this could be a huge relief.
10
9
61
5,045
Since unveiling Digit 5, Agility has announced four new customers: AGCO, which signed a pilot deal, and three unnamed companies.
Now GXO is coming back for more. After 100,000+ tote moves at Flowery Branch, GA, it’s expanding Digit to Statesville, NC, and Lebanon, TN.
Deployments are expected to start in early 2027 with Digit 4, then switch to Digit 5 once it’s commercially available.
Agility’s been on a roll lately, with a Nasdaq listing getting closer.
x.com/XRoboHub/statu…
Now GXO is coming back for more. After 100,000+ tote moves at Flowery Branch, GA, it’s expanding Digit to Statesville, NC, and Lebanon, TN.
Deployments are expected to start in early 2027 with Digit 4, then switch to Digit 5 once it’s commercially available.
Agility’s been on a roll lately, with a Nasdaq listing getting closer.
x.com/XRoboHub/statu…
GXO announces an expanded partnership with Agility. Digit is headed to @GXOLogistics sites in Statesville, NC, and Lebanon, TN, with deployments expected to begin in early 2027 — building on more than 100,000 totes moved at Flowery Branch, GA.
agilityrobotics.com/content/gxo-an…
#Agility #Digit #MadeToWork
9
10
64
3,116
Fourier’s GR-3 looks a lot less cuddly with its covers off 😂
CEO Alex Gu shared this clip from the team’s routine testing: GR-3 climbing up and down six floors autonomously, all in one take.
CEO Alex Gu shared this clip from the team’s routine testing: GR-3 climbing up and down six floors autonomously, all in one take.
Fourier GR-3’s dance moves look so graceful—it’s a totally different vibe from that 'power aesthetic' other robots have. 💃
15
19
178
19,463
RoboHub🤖 retweeted
This ETH Zurich team is pushing humanoid climbing to a new level.
Its Unitree G1 uses one policy for walking, climbing and crawling. The transitions between those movements aren’t manually scripted.
The team trained ACP using 13.2 minutes of human motion data to learn contact patterns, then used reinforcement learning in simulation to adapt to new terrain.
At runtime, onboard sensing and a goal guide where its hands, feet and knees go, without a reference motion to follow.
The team says even the recovery sequence in the ladder demo wasn’t explicitly included in the human demonstrations. x.com/ChongZzZhang/s…
Its Unitree G1 uses one policy for walking, climbing and crawling. The transitions between those movements aren’t manually scripted.
The team trained ACP using 13.2 minutes of human motion data to learn contact patterns, then used reinforcement learning in simulation to adapt to new terrain.
At runtime, onboard sensing and a goal guide where its hands, feet and knees go, without a reference motion to follow.
The team says even the recovery sequence in the ladder demo wasn’t explicitly included in the human demonstrations. x.com/ChongZzZhang/s…
11
21
174
9,253
This looks more fun than a robot duck... a $400 humanoid built around an old phone.
UCLA’s RoMeLa team built PhoneBot from an Android phone and a modular body with 13 motorized joints. It can walk, turn and get up from a face-down position.
The phone supplies the camera, motion power for vision and walking control. The team even ran the core software on a 2017 Honor 9.
The team estimates the hardware cost at around $400, not including the phone. CAD files and the parts list are already online.
UCLA’s RoMeLa team built PhoneBot from an Android phone and a modular body with 13 motorized joints. It can walk, turn and get up from a face-down position.
The phone supplies the camera, motion power for vision and walking control. The team even ran the core software on a 2017 Honor 9.
The team estimates the hardware cost at around $400, not including the phone. CAD files and the parts list are already online.
21
34
286
14,853
Real home data is what robot training needs next.
Maxinsights says it has captured over 2M hours of people doing everyday tasks, with 1.5M hours ready to train on. That includes people cooking and cleaning in real homes, at their normal pace.
There’s a joke that robots can’t do eople teaching them can’t either. Jokes aside, the people who do these chores every day have a lot to teach robots.
The company just raised $30M in seed funding to expand data collection. It says eight of its 30+ customers are on recurring contracts.
Maxinsights says it has captured over 2M hours of people doing everyday tasks, with 1.5M hours ready to train on. That includes people cooking and cleaning in real homes, at their normal pace.
There’s a joke that robots can’t do eople teaching them can’t either. Jokes aside, the people who do these chores every day have a lot to teach robots.
The company just raised $30M in seed funding to expand data collection. It says eight of its 30+ customers are on recurring contracts.
Robots learn by watching people work. Most of what they watch is staged.
At @MaxInsights, we capture real work at natural speed in real homes, restaurants, and hotels. 2M+ hours so far.
Today we’re announcing a $30M seed round led by @peakxvpartners, plus two new products:
MaxLoop: same-day quality scores on every clip
MaxGlove: touch and force data a camera can't capture
Thank you to our investors, builders, operators and customers. More to come.
3
4
23
23,828
More details:
x.com/MaxinsightsAI/…
x.com/MaxinsightsAI/…
We have 2M hours of egocentric human experience recorded. While it's the largest library of datasets, it is also the least informative number we publish.
Two recordings of equal duration can differ by an order of magnitude in what a model can learn from them. One holds k-and-place under fixed conditions. The other holds bimanual manipulation across changing states, with precise geometry, varied contacts, coordinated fingers and tool use. Equal duration, nowhere near equal learning value.
Effective experience ≈ hours × information per hour.
At ten thousand hours that difference is noise. At ten million it is the difference between a corpus and an archive.
0
0
1
459
This ETH Zurich team is pushing humanoid climbing to a new level.
Its Unitree G1 uses one policy for walking, climbing and crawling. The transitions between those movements aren’t manually scripted.
The team trained ACP using 13.2 minutes of human motion data to learn contact patterns, then used reinforcement learning in simulation to adapt to new terrain.
At runtime, onboard sensing and a goal guide where its hands, feet and knees go, without a reference motion to follow.
The team says even the recovery sequence in the ladder demo wasn’t explicitly included in the human demonstrations. x.com/ChongZzZhang/s…
Its Unitree G1 uses one policy for walking, climbing and crawling. The transitions between those movements aren’t manually scripted.
The team trained ACP using 13.2 minutes of human motion data to learn contact patterns, then used reinforcement learning in simulation to adapt to new terrain.
At runtime, onboard sensing and a goal guide where its hands, feet and knees go, without a reference motion to follow.
The team says even the recovery sequence in the ladder demo wasn’t explicitly included in the human demonstrations. x.com/ChongZzZhang/s…
11
21
174
9,253
RoboHub🤖 retweeted
Real home data is what robot training needs next.
Maxinsights says it has captured over 2M hours of people doing everyday tasks, with 1.5M hours ready to train on. That includes people cooking and cleaning in real homes, at their normal pace.
There’s a joke that robots can’t do eople teaching them can’t either. Jokes aside, the people who do these chores every day have a lot to teach robots.
The company just raised $30M in seed funding to expand data collection. It says eight of its 30+ customers are on recurring contracts.
Maxinsights says it has captured over 2M hours of people doing everyday tasks, with 1.5M hours ready to train on. That includes people cooking and cleaning in real homes, at their normal pace.
There’s a joke that robots can’t do eople teaching them can’t either. Jokes aside, the people who do these chores every day have a lot to teach robots.
The company just raised $30M in seed funding to expand data collection. It says eight of its 30+ customers are on recurring contracts.
Robots learn by watching people work. Most of what they watch is staged.
At @MaxInsights, we capture real work at natural speed in real homes, restaurants, and hotels. 2M+ hours so far.
Today we’re announcing a $30M seed round led by @peakxvpartners, plus two new products:
MaxLoop: same-day quality scores on every clip
MaxGlove: touch and force data a camera can't capture
Thank you to our investors, builders, operators and customers. More to come.
3
4
23
23,828
Ropedia has been rolling out new robotics research all week. Its latest, Ropedia-A0, introduces Code-Only-as-Policy, a new paradigm for robot control. It shows clear advantages over model-in-the-loop paradigms, and it generalizes across embodiments and from sim to real.
On aches a 70.24% success rate. Its policy is a shared code library with a thin program for each task. A task program can be reused across episodes to generate any number of trajectories, and skills can be reused across tasks. This makes Ropedia-A0 both an efficient data engine and a stable, reusable robot policy.
The team also shows that the shared library generalizes. It carries over to new embodiments in other datasets like LIBERO and other simulators like MuJoCo, with only a thin adapter for each. From sim to real, it achieves 75% first-try success on pick-and-place, with no retries and no refactoring.
Ropedia-A0 points to a new direction: Code-Only-as-Policy. A generalizable code library that needs no online model can complete tasks efficiently and generate trajectories for model training, opening up exciting potential for robot policy.
On aches a 70.24% success rate. Its policy is a shared code library with a thin program for each task. A task program can be reused across episodes to generate any number of trajectories, and skills can be reused across tasks. This makes Ropedia-A0 both an efficient data engine and a stable, reusable robot policy.
The team also shows that the shared library generalizes. It carries over to new embodiments in other datasets like LIBERO and other simulators like MuJoCo, with only a thin adapter for each. From sim to real, it achieves 75% first-try success on pick-and-place, with no retries and no refactoring.
Ropedia-A0 points to a new direction: Code-Only-as-Policy. A generalizable code library that needs no online model can complete tasks efficiently and generate trajectories for model training, opening up exciting potential for robot policy.
3
7
23
24,349
More on Ropedia-A0 👇
Project page: ropedia.github.io/ropedia-a0/
Paper: arxiv.org/pdf/2610.12369
Hugging Face: huggingface.co/papers/2610.12…
Project page: ropedia.github.io/ropedia-a0/
Paper: arxiv.org/pdf/2610.12369
Hugging Face: huggingface.co/papers/2610.12…
0
0
4
544
38 DoF... YQ Bionics has shared a first look inside the Y-Hand M2, which it calls the world's highest-DoF bionic hand.
Look at the joints. Bone-like segments, cartilage-like components and braided ligaments work together around curved joint surfaces. The hard parts carry oft elements carry tension, giving the fingers compliance built into their structure.
The pull comes from its in-house YQ Motors and YQ Muscles artificial muscle modules. Tendons running through sheaths carry that force across multiple joints, letting several actuators work together to move the fingers.
Encoders and force sensors tell the controller how the hand is moving and what forces it's experiencing. Combined with vision, that feedback helps the fingers conform to an object's shape and maintain their grip.
Look at the joints. Bone-like segments, cartilage-like components and braided ligaments work together around curved joint surfaces. The hard parts carry oft elements carry tension, giving the fingers compliance built into their structure.
The pull comes from its in-house YQ Motors and YQ Muscles artificial muscle modules. Tendons running through sheaths carry that force across multiple joints, letting several actuators work together to move the fingers.
Encoders and force sensors tell the controller how the hand is moving and what forces it's experiencing. Combined with vision, that feedback helps the fingers conform to an object's shape and maintain their grip.
12
37
187
10,046
RoboHub🤖 retweeted
The first Humanoid Hub Conference was incredible.
Pulling off an event like this is far harder than I imagined, and it would not have been possible without the people around me.
Thanks to @itsajchan and the team, who carried a huge amount of this.
Pulling off an event like this is far harder than I imagined, and it would not have been possible without the people around me.
Thanks to @itsajchan and the team, who carried a huge amount of this.
17
8
94
13,384
H2-X has hooks in place of hands for this lift.
That looks like an older Nissan Sylphy, around 1.2 metric tons in stock form.
By that estimate, lifting the front wheels off the ground could put roughly half a metric ton across the two hooks. What do you think?
That looks like an older Nissan Sylphy, around 1.2 metric tons in stock form.
By that estimate, lifting the front wheels off the ground could put roughly half a metric ton across the two hooks. What do you think?
This is wild... 🤯 Unitree just unveiled H2-X, and it’s an absolute beast.
Spinning kicks, aerial flips, powerful knee and elbow strikes. It jumps from a 4-meter platform onto mats, rolls with the impact, and gets back up. That whole sequence is a serious test of joint strength and whole-body control.
The launch video even casts it as a personal bodyguard. It faces off with an intruder and, in a staged rescue, lifts the front of a car while someone pulls a dummy out from underneath.
Unitree says the footage is real and not sped up. It also reports roughly eight hours of battery life while standing.
x.com/UnitreeRobotic…
35
18
120
15,369