Загрузка видео...

Не удалось загрузить видео

На главную

TARS debuted their DexHand at the world's most prestigious robotics conference this week. The hand has 21 degrees of freedom built at 1:1 human anatomical scale. Inside each fingertip: sensors that detect texture at 0.05mm resolution at over 240Hz. A human hair is 0.07mm wide. The hand is more...

heyshrutimishra's profile picture

Shruti

181,819 subscribers

12,979 просмотров • 3 месяцев назад •via X (Twitter)

LIVE

Anya Rossi• Live Now

Private livecam show

Комментарии: 0

Нет доступных комментариев

Здесь появятся комментарии из оригинального поста

Похожие видео

TARS has made some big announcements at ICRA 2026. DexHand - A dexterous hand with 1:1 replication of human metacarpal and finger structure, multi-axis thumb opposition - Quasi-direct-drive + backdrivable joints, 21 degrees of freedom - All 26 ASL letter signs reproduced at high precision -0.05mm fingertip tactile sensing - Data. Model. Hand. Fused into one embodied AI-native dexterous hand. OmniVTA with TacForeSight A visuo-tactile world model for general-use, paired with TacForeSight, a force-guided technology that not only allows robots to understand contact but also enables them to predict contact before it happens. - Semantic touch ("slippery," "rough," "soft," "hard") enabling texture-based material cognition. - Operates inside a real-time closed loop - Contact state prediction and reaction at the microsecond scale, faster than human AWE 3.0 - Claimed to be the world's leading general-purpose embodied foundation model for real-world work - A1 robot packing demo as the showcase: long-horizon, bimanual, with autonomous error recovery and end-to-end imitation learning - Set a Guinness World Record in March 2026 for "Most sub-millimeter wire harness assembled by a robot in one hour" - SenseHub data acquisition suite captures human experience data to train AWE 3.0
0:40

TARS has made some big announcements at ICRA 2026. DexHand - A dexterous hand with 1:1 replication of human metacarpal and finger structure, multi-axis thumb opposition - Quasi-direct-drive + backdrivable joints, 21 degrees of freedom - All 26 ASL letter signs reproduced at high precision -0.05mm fingertip tactile sensing - Data. Model. Hand. Fused into one embodied AI-native dexterous hand. OmniVTA with TacForeSight A visuo-tactile world model for general-use, paired with TacForeSight, a force-guided technology that not only allows robots to understand contact but also enables them to predict contact before it happens. - Semantic touch ("slippery," "rough," "soft," "hard") enabling texture-based material cognition. - Operates inside a real-time closed loop - Contact state prediction and reaction at the microsecond scale, faster than human AWE 3.0 - Claimed to be the world's leading general-purpose embodied foundation model for real-world work - A1 robot packing demo as the showcase: long-horizon, bimanual, with autonomous error recovery and end-to-end imitation learning - Set a Guinness World Record in March 2026 for "Most sub-millimeter wire harness assembled by a robot in one hour" - SenseHub data acquisition suite captures human experience data to train AWE 3.0

The Humanoid Hub

11,322 просмотров • 3 месяцев назад

Reward AI just came out of stealth and unveiled OM-1, a robot policy learned entirely from humans. No teleoperation, no on-robot data. One Model, Any Body. 1. Capture. A wearable 7-DoF hand (Omnibody Hand) lets people demonstrate manipulation naturally while working, cooking, or sorting. Sensors record tactile, proximity, force and hand pose at full human speed. 2. Learn. OM-1 trains on this human data alone. No teleop, no on-robot experience. Then it runs across bodies from industrial arms to humanoids, and scaling is just adding more human data. 3. Control. A high-frequency control layer, trained with RL in sim, executes those actions on any robot, holding steady through unknown loads like a stuck fridge door. The result: OM-1 picks up a brand-new task from under 30 minutes of data.
0:27

Reward AI just came out of stealth and unveiled OM-1, a robot policy learned entirely from humans. No teleoperation, no on-robot data. One Model, Any Body. 1. Capture. A wearable 7-DoF hand (Omnibody Hand) lets people demonstrate manipulation naturally while working, cooking, or sorting. Sensors record tactile, proximity, force and hand pose at full human speed. 2. Learn. OM-1 trains on this human data alone. No teleop, no on-robot experience. Then it runs across bodies from industrial arms to humanoids, and scaling is just adding more human data. 3. Control. A high-frequency control layer, trained with RL in sim, executes those actions on any robot, holding steady through unknown loads like a stuck fridge door. The result: OM-1 picks up a brand-new task from under 30 minutes of data.

The Humanoid Hub

35,477 просмотров • 15 дней назад

when we were at facebook, we believed that at some point in the future, most of the transactions on the internet would not be done by humans they would be done by machines that conviction shaped every architectural decision behind sui we built sui for the world we knew was coming a world where machines will be the primary economic actors on the internet and that world is no longer a forecast. it is unfolding right in front of us the internet has reached a tipping point where automated activity, supercharged by AI, now outpaces human interaction non-human traffic now accounts for more than 50% of all global web activity and you can see humans using agentic workflows more and more in their daily lives in the next years, that trend is going to grow exponentially and the volume of financial transactions executed by agents is also going to grow exponentially with it each agentic workload will be running multiple thousand economic transactions a second and this is going to be orders of magnitude higher than what human wallets do today the L1s optimized for human usage patterns, human attention, human accounts, and human patience cannot adapt to where this is going i have always said this if it is not in the foundation, you cannot patch your way to it later and rn, no other L1 has the foundation sui has this is why agentic apps like Beep, Audric, WaterX are choosing sui and this is just a start. more agentic apps will keep landing on sui because agents are optimizers. they will always route through the fastest, cheapest path on the internet and that path is sui we believed it at facebook. we believe it more today than we ever did the agentic economy is inevitable. and it will run on Sui
2:44

when we were at facebook, we believed that at some point in the future, most of the transactions on the internet would not be done by humans they would be done by machines that conviction shaped every architectural decision behind sui we built sui for the world we knew was coming a world where machines will be the primary economic actors on the internet and that world is no longer a forecast. it is unfolding right in front of us the internet has reached a tipping point where automated activity, supercharged by AI, now outpaces human interaction non-human traffic now accounts for more than 50% of all global web activity and you can see humans using agentic workflows more and more in their daily lives in the next years, that trend is going to grow exponentially and the volume of financial transactions executed by agents is also going to grow exponentially with it each agentic workload will be running multiple thousand economic transactions a second and this is going to be orders of magnitude higher than what human wallets do today the L1s optimized for human usage patterns, human attention, human accounts, and human patience cannot adapt to where this is going i have always said this if it is not in the foundation, you cannot patch your way to it later and rn, no other L1 has the foundation sui has this is why agentic apps like Beep, Audric, WaterX are choosing sui and this is just a start. more agentic apps will keep landing on sui because agents are optimizers. they will always route through the fastest, cheapest path on the internet and that path is sui we believed it at facebook. we believe it more today than we ever did the agentic economy is inevitable. and it will run on Sui

Adeniyi.sui

24,920 просмотров • 4 месяцев назад

Holy…S😳 Xiaomi's New CyberOne is so human-like Although this update features a bionic hand, I was immediately drawn to it. Let's look at the changes in the hand： It can handle industrial precision tasks like turning screws, plus delicate operations such as pinching feathers and throwing balloons. Behind the performance: >Volume cut by 60%:now almost identical in size/shape to a real human hand >Big leap in degrees of freedom (+50% total, +83% active:22-27DOF) >Full-palm tactile sensors over 8200 mm² for precise grip even without vision >150,000+ grip cycles durability (61-hour test) And a major innovation:Smart bionic sweat gland cooling: evaporates water for ~10W active heat dissipation Using tactile gloves to capture real human data, they’re training smoother, human-like grasps with imitation and reinforcement learning. Elon has said that humanoid hands and true AI are the most difficult aspects of building humanoid robots. It seems that Xiaomi is also getting close.
0:41

Holy…S😳 Xiaomi's New CyberOne is so human-like Although this update features a bionic hand, I was immediately drawn to it. Let's look at the changes in the hand： It can handle industrial precision tasks like turning screws, plus delicate operations such as pinching feathers and throwing balloons. Behind the performance: >Volume cut by 60%:now almost identical in size/shape to a real human hand >Big leap in degrees of freedom (+50% total, +83% active:22-27DOF) >Full-palm tactile sensors over 8200 mm² for precise grip even without vision >150,000+ grip cycles durability (61-hour test) And a major innovation:Smart bionic sweat gland cooling: evaporates water for ~10W active heat dissipation Using tactile gloves to capture real human data, they’re training smoother, human-like grasps with imitation and reinforcement learning. Elon has said that humanoid hands and true AI are the most difficult aspects of building humanoid robots. It seems that Xiaomi is also getting close.

CyberRobo

111,430 просмотров • 6 месяцев назад

I need you to sit down for a moment and fully understand this: THE COUNTRY THAT SOLVES AI AND ROBOTICS WILL RULE THE WORLD AND SPACE. The west really only has Elon Musk, the east has 100s of companies fortified by an ENTIRE COUNTRY. Now some will argue no this is not true. No US company has the scale to MANUFACTURE, the compute power, and the finances to compete. Just a few hours ago we saw IRON for the first time, now you will. IRON, a 5’10, 150-lb AI humanoid robots are already building EV cars on the XPENG Motors factory floor. It has over 60 joints, a human-like spine, facial expressions, and male/female customizations The gait of this robot is the most human-like ever seen. Mass rollout in 2026. It is a very big deal. Because as the west does the best in clubbing each other over its head the last decade, China has looked and laughed and built at scale with a fortified government that has little diversion of goal, a 1000 year plan. The west has quietly plans and layers of lawyers and politicians. This is about where YOU LIVE and how you want to live. So when we kick the one person that is Atlas carrying our chance, in the groin, you make a choice on who’s world view you want. It is that simple. No, it is that simple. It ain’t no iPhone it is: whose’s world view will sustain. I can say no one is ready for what I have seen that is up for the next few years. You will think my bombastics were too tame.
7:10

I need you to sit down for a moment and fully understand this: THE COUNTRY THAT SOLVES AI AND ROBOTICS WILL RULE THE WORLD AND SPACE. The west really only has Elon Musk, the east has 100s of companies fortified by an ENTIRE COUNTRY. Now some will argue no this is not true. No US company has the scale to MANUFACTURE, the compute power, and the finances to compete. Just a few hours ago we saw IRON for the first time, now you will. IRON, a 5’10, 150-lb AI humanoid robots are already building EV cars on the XPENG Motors factory floor. It has over 60 joints, a human-like spine, facial expressions, and male/female customizations The gait of this robot is the most human-like ever seen. Mass rollout in 2026. It is a very big deal. Because as the west does the best in clubbing each other over its head the last decade, China has looked and laughed and built at scale with a fortified government that has little diversion of goal, a 1000 year plan. The west has quietly plans and layers of lawyers and politicians. This is about where YOU LIVE and how you want to live. So when we kick the one person that is Atlas carrying our chance, in the groin, you make a choice on who’s world view you want. It is that simple. No, it is that simple. It ain’t no iPhone it is: whose’s world view will sustain. I can say no one is ready for what I have seen that is up for the next few years. You will think my bombastics were too tame.

Brian Roemmele

446,861 просмотров • 10 месяцев назад

A jaw-on-the-floor moment for the whole team was when we had taught the robot to pick up a baggie and shake a usb-brick out with the left hand. During one rollout, it decided to do it with the right hand—and we all stopped. We were stunned, because no one taught it to do that. This was one of the first models with our pretrained base, and it was the moment we realized there was something here that was very different from what any of us had ever built before. We now know this to be part of the improvisational intelligence that emerges with large-scale pretraining on real manipulation data, and it continues to be one of the things that we believe to be worth pushing forward as capability. Our CTO and cofounder Andy Barry joined Association for Advancing Automation to share more about these moments, and how his time at Boston Dynamics and the Broad Institute drove him to focus on building the 🧠 for robots at Generalist. Full ep 👇
0:28

A jaw-on-the-floor moment for the whole team was when we had taught the robot to pick up a baggie and shake a usb-brick out with the left hand. During one rollout, it decided to do it with the right hand—and we all stopped. We were stunned, because no one taught it to do that. This was one of the first models with our pretrained base, and it was the moment we realized there was something here that was very different from what any of us had ever built before. We now know this to be part of the improvisational intelligence that emerges with large-scale pretraining on real manipulation data, and it continues to be one of the things that we believe to be worth pushing forward as capability. Our CTO and cofounder Andy Barry joined Association for Advancing Automation to share more about these moments, and how his time at Boston Dynamics and the Broad Institute drove him to focus on building the 🧠 for robots at Generalist. Full ep 👇

Generalist

11,117 просмотров • 3 месяцев назад

33 years of human work went into mapping a brain smaller than a poppy seed. This is all of it, every one of its 139,255 neurons. 1 female fruit fly was sliced into 7,050 layers, each thinner than a virus, then shot in 21 million photos. AI traced the cells, and humans checked every wire by hand: scientists, students and gamers clicking for fun. The final map: 139,255 neurons 54.5M synapses 8,453 cell types 150 meters of wire folded into 1 speck No one had mapped a full adult brain before, not a mouse and not a human. A fly went first. In September 2026, Janelia and Google released the male, with 166,700 neurons mapped from brain to legs through an intact neck. 2 weeks later, the internet put the fly to work: It played Doom It flew a simulated drone with zero training It landed a simulated rocket 60 times out of 80 It wrote LinkedIn posts But the map only shows where the wires go, not how strong each one is. It has no voltages and no timing, like a circuit diagram with no values on it. Nobody knows what the fly is thinking, and it posts on LinkedIn anyway. It took 33 person-years to map 1 fly, and your head holds 617,000 of them.
0:24

33 years of human work went into mapping a brain smaller than a poppy seed. This is all of it, every one of its 139,255 neurons. 1 female fruit fly was sliced into 7,050 layers, each thinner than a virus, then shot in 21 million photos. AI traced the cells, and humans checked every wire by hand: scientists, students and gamers clicking for fun. The final map: 139,255 neurons 54.5M synapses 8,453 cell types 150 meters of wire folded into 1 speck No one had mapped a full adult brain before, not a mouse and not a human. A fly went first. In September 2026, Janelia and Google released the male, with 166,700 neurons mapped from brain to legs through an intact neck. 2 weeks later, the internet put the fly to work: It played Doom It flew a simulated drone with zero training It landed a simulated rocket 60 times out of 80 It wrote LinkedIn posts But the map only shows where the wires go, not how strong each one is. It has no voltages and no timing, like a circuit diagram with no values on it. Nobody knows what the fly is thinking, and it posts on LinkedIn anyway. It took 33 person-years to map 1 fly, and your head holds 617,000 of them.

Spike 1%

483,928 просмотров • 6 дней назад

🚨 SOUTH DAKOTA COUPLE WAKES UP TO BARE HUMAN FOOTPRINTS AT THEIR FRONT DOOR — BUT THE FEET ARE FACING OPPOSITE DIRECTIONS After a fresh snowfall, they find a single set of bare human footprints coming straight out of the woods… right up to their house. No vehicle marks. No sign of anyone nearby. But it gets worse. Each step is wrong. One footprint points TOWARD the house. The other points AWAY from it. Like someone walked in two directions at once. She follows the trail into the tree line. It keeps going. Halfway through, one of the prints starts to look less human… almost like a paw. Then suddenly they just stop. And remember, this trail started at their front door. What the hell walks barefoot to your door… with its feet reversed... and then disappears?
2:44

🚨 SOUTH DAKOTA COUPLE WAKES UP TO BARE HUMAN FOOTPRINTS AT THEIR FRONT DOOR — BUT THE FEET ARE FACING OPPOSITE DIRECTIONS After a fresh snowfall, they find a single set of bare human footprints coming straight out of the woods… right up to their house. No vehicle marks. No sign of anyone nearby. But it gets worse. Each step is wrong. One footprint points TOWARD the house. The other points AWAY from it. Like someone walked in two directions at once. She follows the trail into the tree line. It keeps going. Halfway through, one of the prints starts to look less human… almost like a paw. Then suddenly they just stop. And remember, this trail started at their front door. What the hell walks barefoot to your door… with its feet reversed... and then disappears?

HustleBitch

2,732,590 просмотров • 7 месяцев назад

.Andrew Yang🧢⬆️🇺🇸 says AI will crush millions of jobs. Theo Jaffee argues jobs don’t work that way and 200 years of automation says otherwise: Andrew Yang: "2.9 million Americans work in customer service call centers. 2 million drive an Uber. 70 million Americans work at white collar jobs. I think AI is going to crush this number." Theo: "Just looking at the past 200 years or so of automation, this claim has been made over and over again and it doesn't stand because automating jobs is really hard." "Jobs are bundles of tasks. In order to automate a job, you have to do every single task the human does on some Pareto curve of better and cheaper than the human. If there's one task in a job remaining for humans to do, humans are still employable" Andrew Yang: "There are 100 people in a department doing a job and if you can reduce 60% of the tasks, what happens is you fire 60 of the hundred and the remaining 40 you say, hey guys, your jobs just changed. Then you systematically reduce that number." Jubilee
3:22

.Andrew Yang🧢⬆️🇺🇸 says AI will crush millions of jobs. Theo Jaffee argues jobs don’t work that way and 200 years of automation says otherwise: Andrew Yang: "2.9 million Americans work in customer service call centers. 2 million drive an Uber. 70 million Americans work at white collar jobs. I think AI is going to crush this number." Theo: "Just looking at the past 200 years or so of automation, this claim has been made over and over again and it doesn't stand because automating jobs is really hard." "Jobs are bundles of tasks. In order to automate a job, you have to do every single task the human does on some Pareto curve of better and cheaper than the human. If there's one task in a job remaining for humans to do, humans are still employable" Andrew Yang: "There are 100 people in a department doing a job and if you can reduce 60% of the tasks, what happens is you fire 60 of the hundred and the remaining 40 you say, hey guys, your jobs just changed. Then you systematically reduce that number." Jubilee

MTS

34,918 просмотров • 2 дней назад

Larry Ellison just called AI the biggest thing in human history. He was being conservative. Ellison: “It is a much bigger deal than the industrial revolution, electricity, whatever. Everything that’s come before.” “Bigger” is the wrong word. “Bigger” assumes the same axis. A taller building on the same foundation. A faster car on the same road. This isn’t a taller building. This is a new dimension. Every revolution in human history extended the body. Steam replaced muscle. Electricity replaced fire. The combustion engine replaced the horse. Each one extraordinary. Each one civilization-altering. Each one built on an assumption nobody ever thought to question. That human cognition was the ceiling. A hammer builds nothing the carpenter can’t envision. A telescope reveals nothing the astronomer can’t interpret. A calculator solves nothing the mathematician can’t frame. 300,000 years of invention. Every tool a servant to the mind that forged it. AI ended that arrangement. Ellison: “We created neural networks that can answer questions that human brains would struggle with.” Not slower. Not less efficiently. Struggle with. We built something that thinks past the point where human thought stops. The tool is no longer bound by the toolmaker. Every ceiling humanity ever hit wasn’t physics. Wasn’t the universe setting limits. It was us. We were the boundary. And we just built something that doesn’t know it’s there. That’s not a bigger industrial revolution. That’s not even a revolution. That’s a species discovering it was the only thing standing between itself and everything it couldn’t yet imagine. And Ellison isn’t waiting for any of it. He’s already past AGI. Already framing superintelligence not as theory. As a scheduling problem. People hear that and reach for fear. Wrong instinct. Every prior revolution displaced labor. AI displaces limits. The industrial revolution didn’t make blacksmiths more creative. It made them irrelevant. AI inverts that. It doesn’t replace human cognition. It uncaps it. The kid with no teachers now has the most patient tutor ever built. The founder with no legal team now has one. The researcher at the edge of what one mind can hold now has a partner with no edge of its own. This isn’t automation. This is cognitive liberation. Every revolution before this answered a human question. This is the first one capable of asking its own. And we haven’t heard the first one yet.
1:17

Larry Ellison just called AI the biggest thing in human history. He was being conservative. Ellison: “It is a much bigger deal than the industrial revolution, electricity, whatever. Everything that’s come before.” “Bigger” is the wrong word. “Bigger” assumes the same axis. A taller building on the same foundation. A faster car on the same road. This isn’t a taller building. This is a new dimension. Every revolution in human history extended the body. Steam replaced muscle. Electricity replaced fire. The combustion engine replaced the horse. Each one extraordinary. Each one civilization-altering. Each one built on an assumption nobody ever thought to question. That human cognition was the ceiling. A hammer builds nothing the carpenter can’t envision. A telescope reveals nothing the astronomer can’t interpret. A calculator solves nothing the mathematician can’t frame. 300,000 years of invention. Every tool a servant to the mind that forged it. AI ended that arrangement. Ellison: “We created neural networks that can answer questions that human brains would struggle with.” Not slower. Not less efficiently. Struggle with. We built something that thinks past the point where human thought stops. The tool is no longer bound by the toolmaker. Every ceiling humanity ever hit wasn’t physics. Wasn’t the universe setting limits. It was us. We were the boundary. And we just built something that doesn’t know it’s there. That’s not a bigger industrial revolution. That’s not even a revolution. That’s a species discovering it was the only thing standing between itself and everything it couldn’t yet imagine. And Ellison isn’t waiting for any of it. He’s already past AGI. Already framing superintelligence not as theory. As a scheduling problem. People hear that and reach for fear. Wrong instinct. Every prior revolution displaced labor. AI displaces limits. The industrial revolution didn’t make blacksmiths more creative. It made them irrelevant. AI inverts that. It doesn’t replace human cognition. It uncaps it. The kid with no teachers now has the most patient tutor ever built. The founder with no legal team now has one. The researcher at the edge of what one mind can hold now has a partner with no edge of its own. This isn’t automation. This is cognitive liberation. Every revolution before this answered a human question. This is the first one capable of asking its own. And we haven’t heard the first one yet.

Dustin

14,852 просмотров • 3 месяцев назад

Ex Machina is no longer sci-fi. China has finally built it. The company is AheadForm, founded in Shanghai. The product is the world's most hyper-realistic robotic face. Silicone skin you can't tell from human, 25 micro motors hidden underneath pulling the face into real expressions. And RGB cameras embedded inside the pupils so when it looks at you, it actually sees you from where its eyes are. They raised $28.5M to "give AI a head," which is also where the name comes from. AheadForm = a head form. This is the opposite of where everyone else in robotics is focused. Unitree, Figure, Tesla, Boston Dynamics: all about the body. AheadForm chose the face because they think trust is the harder problem to solve, and trust gets decided at the face. The reason nobody else has tried this is the "uncanny valley." It's the creepy zone where a robot looks almost human but not quite, and looking at it just feels wrong even when you can't say why. Most roboticists believed no amount of engineering could make a face realistic enough to escape it. So they gave up and kept robots cartoonish on purpose: big anime eyes, exaggerated features, clearly synthetic. But AheadForm decided to treat it as an engineering bug instead. Add enough motors, tune the silicone, fix the timing, the valley closes. And they're pulling it off. A few crazy details about how this actually works: 1. The robot learns its own face in a mirror. You put it in front of a camera, let it fire every motor randomly, and it watches what its face does and builds an internal map of "if I send command X to motor Y, my eyebrow does this." Same exact process a human baby uses staring into a mirror. The robot teaches itself who it is by experimenting. 2. It predicts your smile 839 milliseconds before you smile. By watching the micro-tells in your face that precede a smile, the robot starts smiling 0.8 seconds ahead, so its smile lands at the same moment yours does. Most robot mimicry happens half a second late, which is exactly why it always feels artificial. 3. The pupils are the cameras. When the robot makes eye contact, the gaze and the sensor are the same physical thing. Most humanoid robots stick the camera on the forehead or chest, so they aren't actually looking at you when their eyes are pointed at you. 4. The founder, Yuhang Hu, did his PhD at Columbia under Hod Lipson. Lipson is the guy who in 2006 built a four-legged robot that figured out it had four legs by experimenting with its own movement, nobody told it the body shape, it discovered it. He has spent 25 years trying to build machines that know what they are. AheadForm is that 25-year research arc productized. 5. NetEase Games already paid them to physically embody a fantasy video game character. That opens up a brand-new category: robotics as the physical embodiment of fictional IP. Every character-rich studio, Disney, Riot, Hoyoverse, Pokemon, Netflix, now has a question to answer about when their characters get bodies. AheadForm believes whoever ships the first robot you'd actually want around your family wins. That's the bet behind the most realistic robot face on earth.
1:06

Ex Machina is no longer sci-fi. China has finally built it. The company is AheadForm, founded in Shanghai. The product is the world's most hyper-realistic robotic face. Silicone skin you can't tell from human, 25 micro motors hidden underneath pulling the face into real expressions. And RGB cameras embedded inside the pupils so when it looks at you, it actually sees you from where its eyes are. They raised $28.5M to "give AI a head," which is also where the name comes from. AheadForm = a head form. This is the opposite of where everyone else in robotics is focused. Unitree, Figure, Tesla, Boston Dynamics: all about the body. AheadForm chose the face because they think trust is the harder problem to solve, and trust gets decided at the face. The reason nobody else has tried this is the "uncanny valley." It's the creepy zone where a robot looks almost human but not quite, and looking at it just feels wrong even when you can't say why. Most roboticists believed no amount of engineering could make a face realistic enough to escape it. So they gave up and kept robots cartoonish on purpose: big anime eyes, exaggerated features, clearly synthetic. But AheadForm decided to treat it as an engineering bug instead. Add enough motors, tune the silicone, fix the timing, the valley closes. And they're pulling it off. A few crazy details about how this actually works: 1. The robot learns its own face in a mirror. You put it in front of a camera, let it fire every motor randomly, and it watches what its face does and builds an internal map of "if I send command X to motor Y, my eyebrow does this." Same exact process a human baby uses staring into a mirror. The robot teaches itself who it is by experimenting. 2. It predicts your smile 839 milliseconds before you smile. By watching the micro-tells in your face that precede a smile, the robot starts smiling 0.8 seconds ahead, so its smile lands at the same moment yours does. Most robot mimicry happens half a second late, which is exactly why it always feels artificial. 3. The pupils are the cameras. When the robot makes eye contact, the gaze and the sensor are the same physical thing. Most humanoid robots stick the camera on the forehead or chest, so they aren't actually looking at you when their eyes are pointed at you. 4. The founder, Yuhang Hu, did his PhD at Columbia under Hod Lipson. Lipson is the guy who in 2006 built a four-legged robot that figured out it had four legs by experimenting with its own movement, nobody told it the body shape, it discovered it. He has spent 25 years trying to build machines that know what they are. AheadForm is that 25-year research arc productized. 5. NetEase Games already paid them to physically embody a fantasy video game character. That opens up a brand-new category: robotics as the physical embodiment of fictional IP. Every character-rich studio, Disney, Riot, Hoyoverse, Pokemon, Netflix, now has a question to answer about when their characters get bodies. AheadForm believes whoever ships the first robot you'd actually want around your family wins. That's the bet behind the most realistic robot face on earth.

Ole Lehmann

538,338 просмотров • 4 месяцев назад

This is what happens when you hold onto a tree and never let go as it keeps growing. Giuseppe Penone created It Will Continue to Grow Except at That Point by casting his hand around the trunk of a young tree and then leaving it alone. Over time, the tree grew freely in every direction except where the hand once pressed, preserving a single, permanent indentation. Nothing was cut or reshaped by force—the work emerged through time, growth, and restraint. Long after the hand disappeared, its trace remained in the living wood, a quiet record of how even brief human contact can leave a lasting mark on nature.
0:30

This is what happens when you hold onto a tree and never let go as it keeps growing. Giuseppe Penone created It Will Continue to Grow Except at That Point by casting his hand around the trunk of a young tree and then leaving it alone. Over time, the tree grew freely in every direction except where the hand once pressed, preserving a single, permanent indentation. Nothing was cut or reshaped by force—the work emerged through time, growth, and restraint. Long after the hand disappeared, its trace remained in the living wood, a quiet record of how even brief human contact can leave a lasting mark on nature.

Historic Vids

2,894,436 просмотров • 6 месяцев назад

The robot flipped a pancake nobody taught it! 🥞 Skild AI team assumed pancake flipping had to be somewhere in the training data. So they searched. Millions of hours of pre-training data. Nothing. S1 inferred the whole task from a single human demonstration. That's their new general robot model, built as an in-context learner from the ground up. Every new robot task today starts with days of teleoperation and a fine-tuning run on a specialist policy. S1 skips all of it. Much like a language model, it never updates its weights to learn a new task. The demonstration enters the context window, and the policy uses it to decide what to do next. → Ten-minute tasks it was never trained on, composed from primitives learned in pre-training: a new style of coffee, potting a plant, frying pancakes. → Soil and pots arrived at their office at 8:54 PM. The robot was running the task autonomously by 9:27 PM. → Slide objects away mid-reach, swap them, change the lighting, it still finishes. → The prompt waters a plant with a watering can, but only a cup is available. It uses the cup. It doesn't rigidly replay what it saw, but it recovers from its own errors, and sometimes executes with more precision than the demonstrator, when the human fumbles an egg and makes a mess, S1 performs the same step cleanly. The demonstration is a specification of the goal, and not a trajectory to copy. On unseen tasks after 100K hours of pre-training: language-prompted VLAs reach 9%. Their new model reaches 66%. It's already deploying with industrial partners, with a wider rollout over the coming months. Congrats Deepak Pathak and team behind this! 😮‍💨 🔗 Link to their latest blog: ~~ ♻️ Join the weekly robotics newsletter, and never miss any news →
1:34

The robot flipped a pancake nobody taught it! 🥞 Skild AI team assumed pancake flipping had to be somewhere in the training data. So they searched. Millions of hours of pre-training data. Nothing. S1 inferred the whole task from a single human demonstration. That's their new general robot model, built as an in-context learner from the ground up. Every new robot task today starts with days of teleoperation and a fine-tuning run on a specialist policy. S1 skips all of it. Much like a language model, it never updates its weights to learn a new task. The demonstration enters the context window, and the policy uses it to decide what to do next. → Ten-minute tasks it was never trained on, composed from primitives learned in pre-training: a new style of coffee, potting a plant, frying pancakes. → Soil and pots arrived at their office at 8:54 PM. The robot was running the task autonomously by 9:27 PM. → Slide objects away mid-reach, swap them, change the lighting, it still finishes. → The prompt waters a plant with a watering can, but only a cup is available. It uses the cup. It doesn't rigidly replay what it saw, but it recovers from its own errors, and sometimes executes with more precision than the demonstrator, when the human fumbles an egg and makes a mess, S1 performs the same step cleanly. The demonstration is a specification of the goal, and not a trajectory to copy. On unseen tasks after 100K hours of pre-training: language-prompted VLAs reach 9%. Their new model reaches 66%. It's already deploying with industrial partners, with a wider rollout over the coming months. Congrats Deepak Pathak and team behind this! 😮‍💨 🔗 Link to their latest blog: ~~ ♻️ Join the weekly robotics newsletter, and never miss any news →

Lukas Ziegler

11,406 просмотров • 1 месяц назад

One of the greatest blessings of growing up before iPhones and the internet is that our most spectacularly dumb moments stayed between us and the people who were actually there. No permanent digital record. No strangers watching it on loop years later. No “this will live forever” anxiety hanging over every bad decision. We just got to be young and stupid in private… then eventually laugh about it at reunions when the story finally came out. Watching videos like this makes me genuinely grateful for that era. Embarrassing moments are part of being human. They just weren’t meant to be content. Sometimes the best filter is simply never having the camera pointed at you in the first place.
1:07

One of the greatest blessings of growing up before iPhones and the internet is that our most spectacularly dumb moments stayed between us and the people who were actually there. No permanent digital record. No strangers watching it on loop years later. No “this will live forever” anxiety hanging over every bad decision. We just got to be young and stupid in private… then eventually laugh about it at reunions when the story finally came out. Watching videos like this makes me genuinely grateful for that era. Embarrassing moments are part of being human. They just weren’t meant to be content. Sometimes the best filter is simply never having the camera pointed at you in the first place.

NancyH

14,723 просмотров • 1 месяц назад

I find it wild that Physical Intelligence robotics models can work at all given that they only have a 1 second context length. As the robot is moving around a room and cleaning up a table, it only sees the current timestamp of images from the cameras (plus the text description of its task). I asked Sergey Levine was how this model can work at all. If a human had to execute a many minutes long plan in order to accomplish a physical task, but had her memory reset every second, there's no way she could pull it off But somehow just knowing what state the world is literally in this moment is all you need to know to keep a reasonably complication plan going? Sergey's answer calls on Morevac's Paradox, and I thought it was quite interesting. Full episode out tomorrow!
1:36

I find it wild that Physical Intelligence robotics models can work at all given that they only have a 1 second context length. As the robot is moving around a room and cleaning up a table, it only sees the current timestamp of images from the cameras (plus the text description of its task). I asked Sergey Levine was how this model can work at all. If a human had to execute a many minutes long plan in order to accomplish a physical task, but had her memory reset every second, there's no way she could pull it off But somehow just knowing what state the world is literally in this moment is all you need to know to keep a reasonably complication plan going? Sergey's answer calls on Morevac's Paradox, and I thought it was quite interesting. Full episode out tomorrow!

Dwarkesh Patel

126,193 просмотров • 1 год назад

NVIDIA just dropped a crazy new research idea. Get ready. The same next-token trick behind ChatGPT got pointed at a ragdoll. Honestly the result is unsettling. No exotic new architecture. Chop human motion into a vocabulary of tiny movement tokens. Train a plain GPT to guess the next one, decode it into muscle commands, let physics run. Then the strange part shows up. Shove the character and it catches its own step. Knock it flat and it plants a hand and gets back up. NOBODY wrote a reward for standing up. Here is why this rattles robotics people. Every humanoid team right now is buried under hand-tuned controllers, one for each little skill. 🔵 the old way: script every skill and its reward by hand 🔵 this way: pretrain ONE body on a mountain of motion and the instincts come free Falling and getting up was never programmed. It emerged. Movement is just a language you can autocomplete.
1:45

NVIDIA just dropped a crazy new research idea. Get ready. The same next-token trick behind ChatGPT got pointed at a ragdoll. Honestly the result is unsettling. No exotic new architecture. Chop human motion into a vocabulary of tiny movement tokens. Train a plain GPT to guess the next one, decode it into muscle commands, let physics run. Then the strange part shows up. Shove the character and it catches its own step. Knock it flat and it plants a hand and gets back up. NOBODY wrote a reward for standing up. Here is why this rattles robotics people. Every humanoid team right now is buried under hand-tuned controllers, one for each little skill. 🔵 the old way: script every skill and its reward by hand 🔵 this way: pretrain ONE body on a mountain of motion and the instincts come free Falling and getting up was never programmed. It emerged. Movement is just a language you can autocomplete.

clankr

126,066 просмотров • 3 месяцев назад

Harness vs. Graphs, clearly explained! a harness is great, and most people think it is the whole thing: retries, timeouts, a sandbox, a log, the context it assembles before every call. all of that is real work, and all of it wraps exactly one call. run it a hundred times and you have one call, made very safely, a hundred times. Graph engineering fixes this by moving the decision up a layer: not how safely one call is made, but which calls exist to be made at all. you need both, and here is the sentence that resolves the whole confusion: the harness is everything around one call. the graph is everything between them. ↳ around one call: retry, timeout, sandbox, log, assemble the context, hand back a result ↳ between calls: split, fan out, merge, gate, send back Prompts → Context → Harness → Loops → Graphs the harness does not go away when you build a graph. it moves under each node, and now there are five of them, each wrapping a call you would never have made by hand. the trick is knowing which layer a failure belongs to. turn a piece off and run it again. if the call still works, it was the harness. if the wrong step runs at all, it was the graph. people spend weeks hardening a harness around a node that should not have existed. one thing to know before you scale it. most of what people call their agent is a harness with a chat box on it. ↳ it retries, it times out, it logs, it assembles context, it holds one call up beautifully ↳ it has never once decided that a second call should exist, and that is the entire difference that last one catches careful people. a harness that never fails is not evidence the system is right. it is evidence one call went well, which is the smallest possible claim. and the one that eats whole nights: a harness cannot save you from the wrong step running. you can retry a bad decision three times with a clean log and perfect isolation, and all you bought was three copies of it. below i have quoted my full guide on graph engineering. it covers the three topologies, the verifier patterns, and where the gate should actually open. save this and read it below ↓
0:53

Harness vs. Graphs, clearly explained! a harness is great, and most people think it is the whole thing: retries, timeouts, a sandbox, a log, the context it assembles before every call. all of that is real work, and all of it wraps exactly one call. run it a hundred times and you have one call, made very safely, a hundred times. Graph engineering fixes this by moving the decision up a layer: not how safely one call is made, but which calls exist to be made at all. you need both, and here is the sentence that resolves the whole confusion: the harness is everything around one call. the graph is everything between them. ↳ around one call: retry, timeout, sandbox, log, assemble the context, hand back a result ↳ between calls: split, fan out, merge, gate, send back Prompts → Context → Harness → Loops → Graphs the harness does not go away when you build a graph. it moves under each node, and now there are five of them, each wrapping a call you would never have made by hand. the trick is knowing which layer a failure belongs to. turn a piece off and run it again. if the call still works, it was the harness. if the wrong step runs at all, it was the graph. people spend weeks hardening a harness around a node that should not have existed. one thing to know before you scale it. most of what people call their agent is a harness with a chat box on it. ↳ it retries, it times out, it logs, it assembles context, it holds one call up beautifully ↳ it has never once decided that a second call should exist, and that is the entire difference that last one catches careful people. a harness that never fails is not evidence the system is right. it is evidence one call went well, which is the smallest possible claim. and the one that eats whole nights: a harness cannot save you from the wrong step running. you can retry a bad decision three times with a clean log and perfect isolation, and all you bought was three copies of it. below i have quoted my full guide on graph engineering. it covers the three topologies, the verifier patterns, and where the gate should actually open. save this and read it below ↓

Hanako

51,536 просмотров • 18 дней назад

Holy… is this really not a human? That was my reaction too,I could hardly believe it. This is an ultra-humanlike humanoid robot. 1.73 m tall, 70 kg, with 115 degrees of freedom across the body,36 in the hands alone,and full-body tactile skin. … This is KAI, the first humanoid robot from Shenzhen-based KinetixAI. It’s essentially a 1:1 human replica，except it has no face. I wonder if it could use Aheadform’s M1? that will be interesting. Powered by its KAI World Model and trained on large-scale Engocentric data, it can fold clothes, use tools, handle deliveries and takeout, and assist with childcare,covering a wide range of household tasks. Ok, expectations for Tesla Optimus v3 have been raised once again.
1:26

Holy… is this really not a human? That was my reaction too,I could hardly believe it. This is an ultra-humanlike humanoid robot. 1.73 m tall, 70 kg, with 115 degrees of freedom across the body,36 in the hands alone,and full-body tactile skin. … This is KAI, the first humanoid robot from Shenzhen-based KinetixAI. It’s essentially a 1:1 human replica，except it has no face. I wonder if it could use Aheadform’s M1? that will be interesting. Powered by its KAI World Model and trained on large-scale Engocentric data, it can fold clothes, use tools, handle deliveries and takeout, and assist with childcare,covering a wide range of household tasks. Ok, expectations for Tesla Optimus v3 have been raised once again.

CyberRobo

122,304 просмотров • 5 месяцев назад

.Jason Ma on how Dyna-2 learned to untwist a real water bottle cap out of the box from just 13 minutes of dexterous hand data because the robot hand looks like a human's: "We got a new pair of dexterous hand. Most of our robots in the office have a gripper-like hand, so the gap between that versus human is actually quite large." "We thought, if we have a robot that has hands that look like human, wouldn't the transfer be greater? Right before the release, they were able to collect just 13 minutes of data." "They collected only 13 minutes of data and then post-trained the best Dyna model that trained on one million hours of data. The model is just able to, out of the box, start opening bottle caps on real water bottles." "This is very surprising because typically post-training a robot would take at least several hours of data. For a dexterous hand that's much more complex than a gripper, it would actually be much harder to train. But the experimental results we saw was the opposite. Perhaps because the hand looks more like the human, the transfer was easier." Dyna Robotics
1:16

.Jason Ma on how Dyna-2 learned to untwist a real water bottle cap out of the box from just 13 minutes of dexterous hand data because the robot hand looks like a human's: "We got a new pair of dexterous hand. Most of our robots in the office have a gripper-like hand, so the gap between that versus human is actually quite large." "We thought, if we have a robot that has hands that look like human, wouldn't the transfer be greater? Right before the release, they were able to collect just 13 minutes of data." "They collected only 13 minutes of data and then post-trained the best Dyna model that trained on one million hours of data. The model is just able to, out of the box, start opening bottle caps on real water bottles." "This is very surprising because typically post-training a robot would take at least several hours of data. For a dexterous hand that's much more complex than a gripper, it would actually be much harder to train. But the experimental results we saw was the opposite. Perhaps because the hand looks more like the human, the transfer was easier." Dyna Robotics

MTS

13,743 просмотров • 1 месяц назад