The US humanoid robotics scene spans 22 companies and research teams.
I mapped their headquarters across the country, from Silicon Valley to Texas, Boston and New York.
The list includes:
• Tesla
• Figure AI
• Boston Dynamics
• Apptronik
• Agility Robotics
• 1X on
• Sunday Robotics
• Persona AI
• Fauna Robotics (Amazon)
• Faraday Future
• Dexterity
• Noble Machines
• Rhoda AI
• Genesis AI
• Dexmate
• Robot .com
• Weave Robotics
• Reflex Robotics
• Bear Robotics
• Richtech Robotics
• IHMC
The Bay Area has the largest cluster on this map, with 12 entries. Austin also has a presence, with Tesla and Apptronik based there.
What caught my attention is how many companies are concentrated around Silicon Valley, while other teams are spread across Texas, Massachusetts, New York and Oregon.
Which of these companies do you think will lead the US humanoid robotics race?
Which companies have I left out?
I mapped their headquarters across the country, from Silicon Valley to Texas, Boston and New York.
The list includes:
• Tesla
• Figure AI
• Boston Dynamics
• Apptronik
• Agility Robotics
• 1X on
• Sunday Robotics
• Persona AI
• Fauna Robotics (Amazon)
• Faraday Future
• Dexterity
• Noble Machines
• Rhoda AI
• Genesis AI
• Dexmate
• Robot .com
• Weave Robotics
• Reflex Robotics
• Bear Robotics
• Richtech Robotics
• IHMC
The Bay Area has the largest cluster on this map, with 12 entries. Austin also has a presence, with Tesla and Apptronik based there.
What caught my attention is how many companies are concentrated around Silicon Valley, while other teams are spread across Texas, Massachusetts, New York and Oregon.
Which of these companies do you think will lead the US humanoid robotics race?
Which companies have I left out?
🇨🇳 The 30 Chinese leaders in humanoid robotics:
• 1 AgiBot
· 2 Unitree
· 3 UBTECH
· 4 Galbot
· 5 Leju Robotics
· 6 EngineAI
· 7 Fourier
· 8 XPeng
· 9 Robot Era
· 10 Kepler
· 11 LimX Dynamics
· 12 X-Humanoid
· 13 Noetix Robotics
· 14 Astribot
· 15 Dobot
· 16 Xiaomi · 18 Galaxea AI
· 19 AI² Robotics
· 20 PUDU Robotics
· 21 DEEP Robotics
· 22 MagicLab
· 23 GAC
· 24 Chery & AiMOGA
· 25 X Square Robot
· 26 Keenon Robotics
· 27 EX-Robots
· 28 Lingbao CASBOT
· 29 Tencent Robotics X
· 30 Shanghai Humanoid Center
The list isn’t finished yet. I’m going to complete it with another graphic and turn it into a thread.
0
0
1
222
🇨🇳 The 30 Chinese leaders in humanoid robotics:
• 1 AgiBot
· 2 Unitree
· 3 UBTECH
· 4 Galbot
· 5 Leju Robotics
· 6 EngineAI
· 7 Fourier
· 8 XPeng
· 9 Robot Era
· 10 Kepler
· 11 LimX Dynamics
· 12 X-Humanoid
· 13 Noetix Robotics
· 14 Astribot
· 15 Dobot
· 16 Xiaomi · 18 Galaxea AI
· 19 AI² Robotics
· 20 PUDU Robotics
· 21 DEEP Robotics
· 22 MagicLab
· 23 GAC
· 24 Chery & AiMOGA
· 25 X Square Robot
· 26 Keenon Robotics
· 27 EX-Robots
· 28 Lingbao CASBOT
· 29 Tencent Robotics X
· 30 Shanghai Humanoid Center
The list isn’t finished yet. I’m going to complete it with another graphic and turn it into a thread.
• 1 AgiBot
· 2 Unitree
· 3 UBTECH
· 4 Galbot
· 5 Leju Robotics
· 6 EngineAI
· 7 Fourier
· 8 XPeng
· 9 Robot Era
· 10 Kepler
· 11 LimX Dynamics
· 12 X-Humanoid
· 13 Noetix Robotics
· 14 Astribot
· 15 Dobot
· 16 Xiaomi · 18 Galaxea AI
· 19 AI² Robotics
· 20 PUDU Robotics
· 21 DEEP Robotics
· 22 MagicLab
· 23 GAC
· 24 Chery & AiMOGA
· 25 X Square Robot
· 26 Keenon Robotics
· 27 EX-Robots
· 28 Lingbao CASBOT
· 29 Tencent Robotics X
· 30 Shanghai Humanoid Center
The list isn’t finished yet. I’m going to complete it with another graphic and turn it into a thread.
Unpopular opinion: the best home robot might be one that does nothing for you.
Mori, from One Kind Artifact (OKA), won't answer every question, control your smart home or handle household chores. That's part of its design.
It looks like a soft mushroom lamp. Small movements, und, and guided breathing exercises are how it interacts with you.
No long conversations. No list of tasks. Just a quiet presence in the room.
The details:
• Around 30 cm tall
• Camera and microphones
• On-device AI processing, according to OKA
• $649 launch price for the limited Founder Edition
What caught my attention is how little Mori asks from its owner. Most home robots are built around tasks. Mori takes a different approach, focusing on how a robot makes you feel.
Would you spend $649 on a robot for companionship, or would you rather have one that actually does things around the house?
Footage: @onekindartifact
3
1
12
800
Unpopular opinion: the best home robot might be one that does nothing for you.
Mori, from One Kind Artifact (OKA), won't answer every question, control your smart home or handle household chores. That's part of its design.
It looks like a soft mushroom lamp. Small movements, und, and guided breathing exercises are how it interacts with you.
No long conversations. No list of tasks. Just a quiet presence in the room.
The details:
• Around 30 cm tall
• Camera and microphones
• On-device AI processing, according to OKA
• $649 launch price for the limited Founder Edition
What caught my attention is how little Mori asks from its owner. Most home robots are built around tasks. Mori takes a different approach, focusing on how a robot makes you feel.
Would you spend $649 on a robot for companionship, or would you rather have one that actually does things around the house?
Footage: @onekindartifact
Mori, from One Kind Artifact (OKA), won't answer every question, control your smart home or handle household chores. That's part of its design.
It looks like a soft mushroom lamp. Small movements, und, and guided breathing exercises are how it interacts with you.
No long conversations. No list of tasks. Just a quiet presence in the room.
The details:
• Around 30 cm tall
• Camera and microphones
• On-device AI processing, according to OKA
• $649 launch price for the limited Founder Edition
What caught my attention is how little Mori asks from its owner. Most home robots are built around tasks. Mori takes a different approach, focusing on how a robot makes you feel.
Would you spend $649 on a robot for companionship, or would you rather have one that actually does things around the house?
Footage: @onekindartifact
Before you build a robot, watch this.
Here are 18 open source robots,
shown with their 3D models.
1. Microban
2. MiniLoong
3. Poppy Humanoid
4. Otto DIY
5. pib
6. Mini Pupper
7. Petoi Bittle
8. Stanford Doggo
9. SpotMicro
10. OpenQuadruped
11. Puguette
12. SO-ARM100
13. LATOR-X
15. LeKiwi
16. TurtleBot3 Burger
17. OpenBot
18. Duckiebot
These projects cover humanoids, quadrupeds, robotic arms and mobile robots.
Before building, check each project's official repository. Hardware specifications, 3D design files and licences vary between projects.
My personal pick would be SO-ARM100. I'd like to explore its design and see what I could build with it.
Which of these 18 robots would you build first?
3D model credits: Microban (Rhoban, CC BY-NC-SA 4.0) · Poppy Humanoid (Poppy Project, CC BY-SA 4.0) · Otto DIY (ottobot. org, CC BY-SA 4.0) · SpotMicro design by Deok-yeon Kim / KDY0523 (CC BY 3.0) · OpenBot wheel model: FreeCAD library (CC BY 3.0) · Duckiebot model: Duckietown (duckietown. com) · All other models from their respective project repositories.
3
2
14
1,907
I’m seeing more and more people download videos from X and reupload them to X.
It costs nothing to copy the link and add "/video/1" at the end
X already provides this feature, so why download and reupload when you can share the original video link?
It costs nothing to copy the link and add "/video/1" at the end
X already provides this feature, so why download and reupload when you can share the original video link?
0
0
3
474
Before you build a robot, watch this.
Here are 18 open source robots,
shown with their 3D models.
1. Microban
2. MiniLoong
3. Poppy Humanoid
4. Otto DIY
5. pib
6. Mini Pupper
7. Petoi Bittle
8. Stanford Doggo
9. SpotMicro
10. OpenQuadruped
11. Puguette
12. SO-ARM100
13. LATOR-X
15. LeKiwi
16. TurtleBot3 Burger
17. OpenBot
18. Duckiebot
These projects cover humanoids, quadrupeds, robotic arms and mobile robots.
Before building, check each project's official repository. Hardware specifications, 3D design files and licences vary between projects.
My personal pick would be SO-ARM100. I'd like to explore its design and see what I could build with it.
Which of these 18 robots would you build first?
3D model credits: Microban (Rhoban, CC BY-NC-SA 4.0) · Poppy Humanoid (Poppy Project, CC BY-SA 4.0) · Otto DIY (ottobot. org, CC BY-SA 4.0) · SpotMicro design by Deok-yeon Kim / KDY0523 (CC BY 3.0) · OpenBot wheel model: FreeCAD library (CC BY 3.0) · Duckiebot model: Duckietown (duckietown. com) · All other models from their respective project repositories.
Here are 18 open source robots,
shown with their 3D models.
1. Microban
2. MiniLoong
3. Poppy Humanoid
4. Otto DIY
5. pib
6. Mini Pupper
7. Petoi Bittle
8. Stanford Doggo
9. SpotMicro
10. OpenQuadruped
11. Puguette
12. SO-ARM100
13. LATOR-X
15. LeKiwi
16. TurtleBot3 Burger
17. OpenBot
18. Duckiebot
These projects cover humanoids, quadrupeds, robotic arms and mobile robots.
Before building, check each project's official repository. Hardware specifications, 3D design files and licences vary between projects.
My personal pick would be SO-ARM100. I'd like to explore its design and see what I could build with it.
Which of these 18 robots would you build first?
3D model credits: Microban (Rhoban, CC BY-NC-SA 4.0) · Poppy Humanoid (Poppy Project, CC BY-SA 4.0) · Otto DIY (ottobot. org, CC BY-SA 4.0) · SpotMicro design by Deok-yeon Kim / KDY0523 (CC BY 3.0) · OpenBot wheel model: FreeCAD library (CC BY 3.0) · Duckiebot model: Duckietown (duckietown. com) · All other models from their respective project repositories.
Most people miss this one detail about robot datasets: failure is data too.
AGIBOT WORLD 2026 Release 4 focuses on physical reasoning. The robot must track objects handle occlusion respect space constraints and decide what to do next.
The dataset covers 9 reasoning abilities ch to long-horizon planning across everyday tasks like shelf restocking, room cleaning and finding hidden objects.
It includes successful and failed attempts detailed task annotations camera videos robot states and action data collected on the AGIBOT G2.
My observation labeling why a robot failed could be as useful as recording how it succeeded.
4
24
165
8,069
Techniahqrobot | humanoid robots retweeted
Most people miss this one detail about robot datasets: failure is data too.
AGIBOT WORLD 2026 Release 4 focuses on physical reasoning. The robot must track objects handle occlusion respect space constraints and decide what to do next.
The dataset covers 9 reasoning abilities ch to long-horizon planning across everyday tasks like shelf restocking, room cleaning and finding hidden objects.
It includes successful and failed attempts detailed task annotations camera videos robot states and action data collected on the AGIBOT G2.
My observation labeling why a robot failed could be as useful as recording how it succeeded.
AGIBOT WORLD 2026 Release 4 focuses on physical reasoning. The robot must track objects handle occlusion respect space constraints and decide what to do next.
The dataset covers 9 reasoning abilities ch to long-horizon planning across everyday tasks like shelf restocking, room cleaning and finding hidden objects.
It includes successful and failed attempts detailed task annotations camera videos robot states and action data collected on the AGIBOT G2.
My observation labeling why a robot failed could be as useful as recording how it succeeded.
Stop throwing away your old phone. It could still be useful.
Here's why.
1
1
12
7,685
Most people miss this one detail about robot datasets: failure is data too.
AGIBOT WORLD 2026 Release 4 focuses on physical reasoning. The robot must track objects handle occlusion respect space constraints and decide what to do next.
The dataset covers 9 reasoning abilities ch to long-horizon planning across everyday tasks like shelf restocking, room cleaning and finding hidden objects.
It includes successful and failed attempts detailed task annotations camera videos robot states and action data collected on the AGIBOT G2.
My observation labeling why a robot failed could be as useful as recording how it succeeded.
AGIBOT WORLD 2026 Release 4 focuses on physical reasoning. The robot must track objects handle occlusion respect space constraints and decide what to do next.
The dataset covers 9 reasoning abilities ch to long-horizon planning across everyday tasks like shelf restocking, room cleaning and finding hidden objects.
It includes successful and failed attempts detailed task annotations camera videos robot states and action data collected on the AGIBOT G2.
My observation labeling why a robot failed could be as useful as recording how it succeeded.
Stop throwing away your old phone. It could still be useful.
Here's why.
1
1
12
7,685
Stop throwing away your old phone. It could still be useful.
Here's why.
Here's why.
Why do robots have to look like humans?
Meet EELS a 4-meter 100 kg snake like robot developed by NASA JPL to explore terrain where wheels and legs struggle.
It uses rotating screw threads to move across snow sand and ice while lidar and stereo cameras help it map its inal goal? Explore the icy moon Enceladus in search of signs of life.
EELS is still a prototype but it raises an interesting question are we designing robots for how humans move or for what the environment demands?
9
44
324
16,682
This is next level.
Claude turning a simple prompt into a fully interactive 3D office with live Tavus PALs at every desk a Corvette racing game with a PAL sidekick Griffstation mini games and all those easter eggs is pure magic.
The level of polish and playfulness is insane.
Claude turning a simple prompt into a fully interactive 3D office with live Tavus PALs at every desk a Corvette racing game with a PAL sidekick Griffstation mini games and all those easter eggs is pure magic.
The level of polish and playfulness is insane.
I asked Claude to build an office for the @tavus PALs...
I was not expecting it to build THAT kind of office
2
3
13
1,080
Why do robots have to look like humans?
Meet EELS a 4-meter 100 kg snake like robot developed by NASA JPL to explore terrain where wheels and legs struggle.
It uses rotating screw threads to move across snow sand and ice while lidar and stereo cameras help it map its inal goal? Explore the icy moon Enceladus in search of signs of life.
EELS is still a prototype but it raises an interesting question are we designing robots for how humans move or for what the environment demands?
Meet EELS a 4-meter 100 kg snake like robot developed by NASA JPL to explore terrain where wheels and legs struggle.
It uses rotating screw threads to move across snow sand and ice while lidar and stereo cameras help it map its inal goal? Explore the icy moon Enceladus in search of signs of life.
EELS is still a prototype but it raises an interesting question are we designing robots for how humans move or for what the environment demands?
There's one thing 99% of people don't know about openArm 2.0
3
2
32
16,639
There's one thing 99% of people don't know about openArm 2.0
What if a robot arm could let you feel what it touches?
That is the idea behind bilateral teleoperation on OpenArm 2.0, the open-source humanoid arm developed by Enactic.
How it works:
1. A human moves a leader arm.
2. The follower arm copies the motion, joint by joint.
3. resistance, its motors register the extra load.
4. That force is sent back to the leader, so the operator feels the contact and adjusts.
Why it matters:
Many robot learning methods start with human demonstrations recorded by teleoperation. If the operator cannot feel contact, they push too hard, crush objects or stop too early.
Force feedback gives gentler more precise demonstrations which means better training data for imitation learning.
Key specs:
7 joints per arm: shoulder elbow and wrist
A camera inside each gripper for close-range vision
4.1 kg nominal payload per arm, gripper included
6 kg peak payload for short durations
Full mechanical design available: STEP, STL and URDF files
MuJoCo simulation models included
Reality check:
Open hardware is not plug and play. Autonomous manipulation still needs control software, calibration and a lot of testing.
Would you build one to collect your own training data?
0
6
91
7,486
Techniahqrobot | humanoid robots retweeted
What if a robot arm could let you feel what it touches?
That is the idea behind bilateral teleoperation on OpenArm 2.0, the open-source humanoid arm developed by Enactic.
How it works:
1. A human moves a leader arm.
2. The follower arm copies the motion, joint by joint.
3. resistance, its motors register the extra load.
4. That force is sent back to the leader, so the operator feels the contact and adjusts.
Why it matters:
Many robot learning methods start with human demonstrations recorded by teleoperation. If the operator cannot feel contact, they push too hard, crush objects or stop too early.
Force feedback gives gentler more precise demonstrations which means better training data for imitation learning.
Key specs:
7 joints per arm: shoulder elbow and wrist
A camera inside each gripper for close-range vision
4.1 kg nominal payload per arm, gripper included
6 kg peak payload for short durations
Full mechanical design available: STEP, STL and URDF files
MuJoCo simulation models included
Reality check:
Open hardware is not plug and play. Autonomous manipulation still needs control software, calibration and a lot of testing.
Would you build one to collect your own training data?
That is the idea behind bilateral teleoperation on OpenArm 2.0, the open-source humanoid arm developed by Enactic.
How it works:
1. A human moves a leader arm.
2. The follower arm copies the motion, joint by joint.
3. resistance, its motors register the extra load.
4. That force is sent back to the leader, so the operator feels the contact and adjusts.
Why it matters:
Many robot learning methods start with human demonstrations recorded by teleoperation. If the operator cannot feel contact, they push too hard, crush objects or stop too early.
Force feedback gives gentler more precise demonstrations which means better training data for imitation learning.
Key specs:
7 joints per arm: shoulder elbow and wrist
A camera inside each gripper for close-range vision
4.1 kg nominal payload per arm, gripper included
6 kg peak payload for short durations
Full mechanical design available: STEP, STL and URDF files
MuJoCo simulation models included
Reality check:
Open hardware is not plug and play. Autonomous manipulation still needs control software, calibration and a lot of testing.
Would you build one to collect your own training data?
Let me say this again: how can a robot learn to grip an object if it can’t sense how much force it’s applying?
A camera can see a hand closing around a glass, but it can’t directly measure grip force or detect the first signs of slipping.
That’s why MaxGlove’s force data matters for robot learning, with a target of 1 million hours by 2027.
Congratulations for the $30M Seed +
3
2
36
9,954
What if a robot arm could let you feel what it touches?
That is the idea behind bilateral teleoperation on OpenArm 2.0, the open-source humanoid arm developed by Enactic.
How it works:
1. A human moves a leader arm.
2. The follower arm copies the motion, joint by joint.
3. resistance, its motors register the extra load.
4. That force is sent back to the leader, so the operator feels the contact and adjusts.
Why it matters:
Many robot learning methods start with human demonstrations recorded by teleoperation. If the operator cannot feel contact, they push too hard, crush objects or stop too early.
Force feedback gives gentler more precise demonstrations which means better training data for imitation learning.
Key specs:
7 joints per arm: shoulder elbow and wrist
A camera inside each gripper for close-range vision
4.1 kg nominal payload per arm, gripper included
6 kg peak payload for short durations
Full mechanical design available: STEP, STL and URDF files
MuJoCo simulation models included
Reality check:
Open hardware is not plug and play. Autonomous manipulation still needs control software, calibration and a lot of testing.
Would you build one to collect your own training data?
That is the idea behind bilateral teleoperation on OpenArm 2.0, the open-source humanoid arm developed by Enactic.
How it works:
1. A human moves a leader arm.
2. The follower arm copies the motion, joint by joint.
3. resistance, its motors register the extra load.
4. That force is sent back to the leader, so the operator feels the contact and adjusts.
Why it matters:
Many robot learning methods start with human demonstrations recorded by teleoperation. If the operator cannot feel contact, they push too hard, crush objects or stop too early.
Force feedback gives gentler more precise demonstrations which means better training data for imitation learning.
Key specs:
7 joints per arm: shoulder elbow and wrist
A camera inside each gripper for close-range vision
4.1 kg nominal payload per arm, gripper included
6 kg peak payload for short durations
Full mechanical design available: STEP, STL and URDF files
MuJoCo simulation models included
Reality check:
Open hardware is not plug and play. Autonomous manipulation still needs control software, calibration and a lot of testing.
Would you build one to collect your own training data?
Let me say this again: how can a robot learn to grip an object if it can’t sense how much force it’s applying?
A camera can see a hand closing around a glass, but it can’t directly measure grip force or detect the first signs of slipping.
That’s why MaxGlove’s force data matters for robot learning, with a target of 1 million hours by 2027.
Congratulations for the $30M Seed +
3
2
36
9,954
The two repositories behind the 3D model used in the video and the poster:
Robot description (URDF file and 3D meshes): github.com/enactic/openar…
MuJoCo simulation model: github.com/enactic/openar…
Robot description (URDF file and 3D meshes): github.com/enactic/openar…
MuJoCo simulation model: github.com/enactic/openar…
0
0
1
299
I think people are completely missing the point behind MicroDuck.
MicroDuck looks like a cute desk toy. The hardware tells a different story.
At just 25 cm tall and around 780 g, this small robotic duck packs a surprising amount of engineering into its body.
Here’s what’s inside:
• 15 motors controlling its legs, head, neck and beak
• or
• 1 GB RAM and 32 GB storage
• A camera for visual input
• An 8-microphone array for audio input
That combination gives MicroDuck the hardware for movement, vision and voice interaction in a compact form.
The robot needs to manage multiple joints while responding to its surroundings.
Small robots like this are also interesting because they bring robotics closer to everyday environments. A compact platform can be easier to place on a desk, carry around and use for experiments than a full-sized humanoid.
Of course, the hardware specs alone don't tell us how well it performs.
Real-world tests would help answer questions about walking stability, voice recognition, response time and battery life.
Still, MicroDuck is a good reminder that you don't need a human-sized robot to explore the engineering behind embodied AI.
Would you keep a robotic duck on your desk if it could move, see and talk to you?
28
148
982
70,220
And robotics is just getting started, with new companies raising over $30 million.
Let me say this again: how can a robot learn to grip an object if it can’t sense how much force it’s applying?
A camera can see a hand closing around a glass, but it can’t directly measure grip force or detect the first signs of slipping.
That’s why MaxGlove’s force data matters for robot learning, with a target of 1 million hours by 2027.
Congratulations for the $30M Seed +
0
0
4
1,168
Let me say this again: how can a robot learn to grip an object if it can’t sense how much force it’s applying?
A camera can see a hand closing around a glass, but it can’t directly measure grip force or detect the first signs of slipping.
That’s why MaxGlove’s force data matters for robot learning, with a target of 1 million hours by 2027.
Congratulations for the $30M Seed +
A camera can see a hand closing around a glass, but it can’t directly measure grip force or detect the first signs of slipping.
That’s why MaxGlove’s force data matters for robot learning, with a target of 1 million hours by 2027.
Congratulations for the $30M Seed +
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
1
12
4,251
Techniahqrobot | humanoid robots retweeted
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.
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.
44
11
111
105,060
Simulation could also help here.
If you don’t have the resources to build and test everything in the real world, you can train robots in many different environments and learn from thousands of trials before deploying them on real hardware.
The challenge is making those skills work in the real world.
If you don’t have the resources to build and test everything in the real world, you can train robots in many different environments and learn from thousands of trials before deploying them on real hardware.
The challenge is making those skills work in the real world.
Culture is the moat.
Who has the most diverse factory in the world?
Talked to a robotics entrepreneur today who explained this process to me while standing out front of the Physical AI event at Fort Mason. Stealth Mode. Building a robot brain that uses reinforcement learning. In both simulator and in real world.
Think about humanoid putting away forks in our kitchens.
You put your forks in a different place than I do.
So maybe the robot opens a cabinet and three drawers before it finds the correct place, he told me. Each try is a failure. Success gets reinforced.
But this will do “perfect” Chinese food.
I once was at the Pearl Hotel in Shanghai. There was a jazz band there that copied American jazz perfectly but it had no soul.
So what if a worker who grew up on China wears the data collection system home and record his grandma cooking food.
And another worker, the same night, was recording his German grandma. And another one recording her Korean grandma. And another her Peruvian grandma. And another his Iranian grandma.
I had the Iranian one living with us for years.
She made the best I have ever had and I have been to the best in Los Angeles and I love my wife’s cooking. And she just entertained 20 on Saturday and everyone loved it.
Soul. Culture.
The robot can’t learn that online. Elon has so many advantages.
But the first one won’t do all that. Is gonna work in the factory.
I predict that in three years Elon announces a consumer one that is lighter, cheaper, more dexterous, for our homes.
But the factory robot will show signs of this “soul.”
Elon called his AI sentient the other day and a lot of people gave him shit about that.
If Elon figures out how to capture grandma then your robot will have soul and be able to do everything.
One last point. A lot of my neighbors have machine shops in their garages.
I built a Thisle sailboat with my neighbor in his garage. Other neighbors do all sorts of different projects in theirs.
A machine shop edition would be a huge seller.
Here diversity, too, matters. In San Jose on Saturday nights there are a club that show off the low rider cars they built.
They have different technologies in their garage than the robot company has in theirs that I met today. And that is different than the guy down the street who tears apart Ford Sprinter vans.
Optimus could help everyone build things. Finish projects that we have been spending years on. Rebuilding an old Harley kind of thing.
This morning another Chinese robot company has a video of it jumping over a car and lifting it up.
Cute.
But what will get me to buy one is what it could do for me in the garage.
Set your RL to learn that!
Takes a certain culture to get what I am saying.
Can I get an “amen” from the builders?
2
0
5
767