正在加载视频...

视频加载失败

返回首页

Let's reverse engineer this demo. You need 3 things: (1) robust hardware and motor designs that treat simulation as first-class citizen; (2) a human motion capture ("mocap") dataset, such as those for film and gaming characters; (3) massively parallel RL training in GPU-accelerated simulation. Last October, our team trained...

DrJimFan's profile picture

Jim Fan

485,756 subscribers

216,139 次观看 • 1 年前 •via X (Twitter)

教育科学技术
LIVE

Anya Rossi• Live Now

Private livecam show

11 条评论

Jim Fan 的头像
Jim Fan1 年前

A 1.5M-parameter neural net that masters whole body motor coordination:

Project Tower | WISHLIST NOW 的头像
Project Tower | WISHLIST NOW2 年前

Who wants to transform into weird creatures? Play our Demo now on STEAM 👇

Bilawal Sidhu 的头像
Bilawal Sidhu1 年前

if this *is* CGI then i wanna know their light estimation and rendering stack :P

The Highly Automated Cat — e/acc ⏩ 的头像
The Highly Automated Cat — e/acc ⏩1 年前

where are they getting the training data for this sort of thing tho? sneaker manufacturers?

Rey Neill 的头像
Rey Neill1 年前

Hips don’t lie

AI Leaks and News 的头像
AI Leaks and News1 年前

Humanoids may be accelerating faster than AI is

kunleforever (FSD Supervisor)😊 的头像
kunleforever (FSD Supervisor)😊1 年前

We are creating our replacements 🤷🏽‍♂️

The Robot Services Exchange 的头像
The Robot Services Exchange1 年前

Hover is cool

R3PL1C8R Drones 的头像
R3PL1C8R Drones1 年前

I believe this can be done once you find the right algo. But why does it look so fake? Is the acceleration/deceleration just so fast that we never see this in live animals/humans? It doesn't seem to interact with the world properly.

Kazioo 的头像
Kazioo1 年前

Is Isaac the only viable option for faster than real-time RL on the GPU? I guess using a game engine that is more CPU dependent (like UE5) will always be relatively slow. Epic seems to plan supporting physical animation training this way, but I'm wondering if it's a dead end in efficiency.

RyanRejoice 的头像
RyanRejoice1 年前

Very impressive. Humanoid Robotics seemed pretty stagnant for years but now everything is coming together and things are really picking up pace... and I meant that mechanically!

相关视频

Tired of teleoperation? One human video → 1,000s of robot demos. (📍GitHub ) Scaling Robot Data Without Dynamics Simulation or Robot Hardware Real2Render2Real (R2R2R) is a new way to scale robot data without physics simulation or hardware. You take a phone scan + a single monocular human demo. It tracks the motion, renders photorealistic scenes, and generates diverse, robot-agnostic trajectories ready for training. > No teleop, no sim, no robot, just a phone and a video > Train VLA models and diffusion policies directly on the output > Supports multiple robot embodiments with kinematic consistency > 1000s of demos in 1/27 the time of real-world collection Thank you, Max Fu, for sharing!! Project: Paper: Code coming soon: It shows that with the right pipeline, you can scale robot learning data without touching a robot. One of the most interesting directions in scalable robotics today. —— Weekly robotics and AI insights. Subscribe free:
0:29

Tired of teleoperation? One human video → 1,000s of robot demos. (📍GitHub ) Scaling Robot Data Without Dynamics Simulation or Robot Hardware Real2Render2Real (R2R2R) is a new way to scale robot data without physics simulation or hardware. You take a phone scan + a single monocular human demo. It tracks the motion, renders photorealistic scenes, and generates diverse, robot-agnostic trajectories ready for training. > No teleop, no sim, no robot, just a phone and a video > Train VLA models and diffusion policies directly on the output > Supports multiple robot embodiments with kinematic consistency > 1000s of demos in 1/27 the time of real-world collection Thank you, Max Fu, for sharing!! Project: Paper: Code coming soon: It shows that with the right pipeline, you can scale robot learning data without touching a robot. One of the most interesting directions in scalable robotics today. —— Weekly robotics and AI insights. Subscribe free:

Ilir Aliu

42,804 次观看 • 5 个月前

I don’t know if we live in a Matrix, but I know for sure that robots will spend most of their lives in simulation. Let machines train machines. I’m excited to introduce DexMimicGen, a massive-scale synthetic data generator that enables a humanoid robot to learn complex skills from only a handful of human demonstrations. Yes, as few as 5! DexMimicGen addresses the biggest pain point in robotics: where do we get data? Unlike with LLMs, where vast amounts of texts are readily available, you cannot simply download motor control signals from the internet. So researchers teleoperate the robots to collect motion data via XR headsets. They have to repeat the same skill over and over and over again, because neural nets are data hungry. This is a very slow and uncomfortable process. At NVIDIA, we believe the majority of high-quality tokens for robot foundation models will come from simulation. What DexMimicGen does is to trade GPU compute time for human time. It takes one motion trajectory from human, and multiplies into 1000s of new trajectories. A robot brain trained on this augmented dataset will generalize far better in the real world. Think of DexMimicGen as a learning signal amplifier. It maps a small dataset to a large (de facto infinite) dataset, using physics simulation in the loop. In this way, we free humans from babysitting the bots all day. The future of robot data is generative. The future of the entire robot learning pipeline will also be generative. 🧵
0:58

I don’t know if we live in a Matrix, but I know for sure that robots will spend most of their lives in simulation. Let machines train machines. I’m excited to introduce DexMimicGen, a massive-scale synthetic data generator that enables a humanoid robot to learn complex skills from only a handful of human demonstrations. Yes, as few as 5! DexMimicGen addresses the biggest pain point in robotics: where do we get data? Unlike with LLMs, where vast amounts of texts are readily available, you cannot simply download motor control signals from the internet. So researchers teleoperate the robots to collect motion data via XR headsets. They have to repeat the same skill over and over and over again, because neural nets are data hungry. This is a very slow and uncomfortable process. At NVIDIA, we believe the majority of high-quality tokens for robot foundation models will come from simulation. What DexMimicGen does is to trade GPU compute time for human time. It takes one motion trajectory from human, and multiplies into 1000s of new trajectories. A robot brain trained on this augmented dataset will generalize far better in the real world. Think of DexMimicGen as a learning signal amplifier. It maps a small dataset to a large (de facto infinite) dataset, using physics simulation in the loop. In this way, we free humans from babysitting the bots all day. The future of robot data is generative. The future of the entire robot learning pipeline will also be generative. 🧵

Jim Fan

165,246 次观看 • 1 年前

Let's reverse engineer Disney's adorable, lifelike robot! I couldn't find a whitepaper, but this is how I think it's trained: 1. The emotional behaviors are curated by Disney animation artists, keyframe by keyframe. But it cannot be "rendered" directly on the robot because it doesn't take into account the complex real-world physics. 2. Reinforcement learning (RL) is a great tool for training low-level robot controllers. RL needs a reward function to optimize, and it's typically a task reward (e.g. walk in a straight line as fast as possible). The problem is that RL doesn't know what counts as "natural behavior", and often produces weird-looking body postures that somehow still maximize the reward. This is a human alignment problem just like ChatGPT. 3. Enters Adversarial Motion Prior (AMP): a technique that learns the human preference by training a classifier on what we consider "emotional & cute". In GAN literature, this is called a discriminator. Disney artists are good at creating such a dataset. You can then add AMP as an auxiliary reward in simulation to nudge the robot towards desired behaviors. AMP was developed by Peng et al. 2021 and Escontrela et al. 2022. 4. Add lots of data augmentation to make the controller robust to physical disturbances. In RL, it's called "domain randomization". This is a very powerful technique that bridges the gap between simulator and reality. Previously, OpenAI used domain randomization to train a 5-finger robot hand to manipulate a Rubik's Cube: IEEE news article gave hints about the pipeline: Finally, praying for world peace 🙏. I hope robotics like this will bring more joy to the world.
0:58

Let's reverse engineer Disney's adorable, lifelike robot! I couldn't find a whitepaper, but this is how I think it's trained: 1. The emotional behaviors are curated by Disney animation artists, keyframe by keyframe. But it cannot be "rendered" directly on the robot because it doesn't take into account the complex real-world physics. 2. Reinforcement learning (RL) is a great tool for training low-level robot controllers. RL needs a reward function to optimize, and it's typically a task reward (e.g. walk in a straight line as fast as possible). The problem is that RL doesn't know what counts as "natural behavior", and often produces weird-looking body postures that somehow still maximize the reward. This is a human alignment problem just like ChatGPT. 3. Enters Adversarial Motion Prior (AMP): a technique that learns the human preference by training a classifier on what we consider "emotional & cute". In GAN literature, this is called a discriminator. Disney artists are good at creating such a dataset. You can then add AMP as an auxiliary reward in simulation to nudge the robot towards desired behaviors. AMP was developed by Peng et al. 2021 and Escontrela et al. 2022. 4. Add lots of data augmentation to make the controller robust to physical disturbances. In RL, it's called "domain randomization". This is a very powerful technique that bridges the gap between simulator and reality. Previously, OpenAI used domain randomization to train a 5-finger robot hand to manipulate a Rubik's Cube: IEEE news article gave hints about the pipeline: Finally, praying for world peace 🙏. I hope robotics like this will bring more joy to the world.

Jim Fan

314,637 次观看 • 2 年前

🚨🇺🇸FIGURE ROBOT LEARNS TO WALK IN HOURS Turns out you can teach a robot to walk—if you’ve got a few years to burn in a physics simulator (or a few hours of GPU time). Figure’s team trained a neural network using reinforcement learning to mimic natural human locomotion, and the wild part is it worked on the first try with real hardware. No tuning. No fiddling. Just plug and strut. Thanks to sim-to-real transfer and some sneaky domain randomization, the robot didn’t just wobble out of the lab—it walked like it had somewhere to be. Next up: running. Not terrifying at all. Source: Figure
0:54

🚨🇺🇸FIGURE ROBOT LEARNS TO WALK IN HOURS Turns out you can teach a robot to walk—if you’ve got a few years to burn in a physics simulator (or a few hours of GPU time). Figure’s team trained a neural network using reinforcement learning to mimic natural human locomotion, and the wild part is it worked on the first try with real hardware. No tuning. No fiddling. Just plug and strut. Thanks to sim-to-real transfer and some sneaky domain randomization, the robot didn’t just wobble out of the lab—it walked like it had somewhere to be. Next up: running. Not terrifying at all. Source: Figure

Mario Nawfal

117,866 次观看 • 1 年前

Exciting updates on Project GR00T! We discover a systematic way to scale up robot data, tackling the most painful pain point in robotics. The idea is simple: human collects demonstration on a real robot, and we multiply that data 1000x or more in simulation. Let’s break it down: 1. We use Apple Vision Pro (yes!!) to give the human operator first person control of the humanoid. Vision Pro parses human hand pose and retargets the motion to the robot hand, all in real time. From the human’s point of view, they are immersed in another body like the Avatar. Teleoperation is slow and time-consuming, but we can afford to collect a small amount of data. 2. We use RoboCasa, a generative simulation framework, to multiply the demonstration data by varying the visual appearance and layout of the environment. In Jensen’s keynote video below, the humanoid is now placing the cup in hundreds of kitchens with a huge diversity of textures, furniture, and object placement. We only have 1 physical kitchen at the GEAR Lab in NVIDIA HQ, but we can conjure up infinite ones in simulation. 3. Finally, we apply MimicGen, a technique to multiply the above data even more by varying the *motion* of the robot. MimicGen generates vast number of new action trajectories based on the original human data, and filters out failed ones (e.g. those that drop the cup) to form a much larger dataset. To sum up, given 1 human trajectory with Vision Pro -> RoboCasa produces N (varying visuals) -> MimicGen further augments to NxM (varying motions). This is the way to trade compute for expensive human data by GPU-accelerated simulation. A while ago, I mentioned that teleoperation is fundamentally not scalable, because we are always limited by 24 hrs/robot/day in the world of atoms. Our new GR00T synthetic data pipeline breaks this barrier in the world of bits. Scaling has been so much fun for LLMs, and it's finally our turn to have fun in robotics! We are building tools to enable everyone in the ecosystem to scale up with us. Links in thread:
1:21

Exciting updates on Project GR00T! We discover a systematic way to scale up robot data, tackling the most painful pain point in robotics. The idea is simple: human collects demonstration on a real robot, and we multiply that data 1000x or more in simulation. Let’s break it down: 1. We use Apple Vision Pro (yes!!) to give the human operator first person control of the humanoid. Vision Pro parses human hand pose and retargets the motion to the robot hand, all in real time. From the human’s point of view, they are immersed in another body like the Avatar. Teleoperation is slow and time-consuming, but we can afford to collect a small amount of data. 2. We use RoboCasa, a generative simulation framework, to multiply the demonstration data by varying the visual appearance and layout of the environment. In Jensen’s keynote video below, the humanoid is now placing the cup in hundreds of kitchens with a huge diversity of textures, furniture, and object placement. We only have 1 physical kitchen at the GEAR Lab in NVIDIA HQ, but we can conjure up infinite ones in simulation. 3. Finally, we apply MimicGen, a technique to multiply the above data even more by varying the *motion* of the robot. MimicGen generates vast number of new action trajectories based on the original human data, and filters out failed ones (e.g. those that drop the cup) to form a much larger dataset. To sum up, given 1 human trajectory with Vision Pro -> RoboCasa produces N (varying visuals) -> MimicGen further augments to NxM (varying motions). This is the way to trade compute for expensive human data by GPU-accelerated simulation. A while ago, I mentioned that teleoperation is fundamentally not scalable, because we are always limited by 24 hrs/robot/day in the world of atoms. Our new GR00T synthetic data pipeline breaks this barrier in the world of bits. Scaling has been so much fun for LLMs, and it's finally our turn to have fun in robotics! We are building tools to enable everyone in the ecosystem to scale up with us. Links in thread:

Jim Fan

364,380 次观看 • 1 年前

We RL'ed humanoid robots to Cristiano Ronaldo, LeBron James, and Kobe Byrant! These are neural nets running on real hardware at our GEAR lab. Most robot demos you see online speed videos up. We actually *slow them down* so you can enjoy the fluid motions. I'm excited to announce "ASAP", a "real2sim2real" model that masters extremely smooth and dynamic motions for humanoid whole body control. We pretrain the robot in simulation first, but there is a notorious "sim2real" gap: it's very difficult for hand-engineered physics equations to match real world dynamics. Our fix is simple: just deploy a pretrained policy on real hardware, collect data, and replay the motion in sim. The replay will obviously have many errors, but that gives a rich signal to compensate for the physics discrepancy. Use another neural net to learn the delta. Basically, we "patch up" a traditional physics engine, so that the robot can experience almost the real world at scale in GPUs. The future is hybrid simulation: combine the power of classical sim engines refined over decades and the uncanny ability of modern NNs to capture a messy world.
0:50

We RL'ed humanoid robots to Cristiano Ronaldo, LeBron James, and Kobe Byrant! These are neural nets running on real hardware at our GEAR lab. Most robot demos you see online speed videos up. We actually *slow them down* so you can enjoy the fluid motions. I'm excited to announce "ASAP", a "real2sim2real" model that masters extremely smooth and dynamic motions for humanoid whole body control. We pretrain the robot in simulation first, but there is a notorious "sim2real" gap: it's very difficult for hand-engineered physics equations to match real world dynamics. Our fix is simple: just deploy a pretrained policy on real hardware, collect data, and replay the motion in sim. The replay will obviously have many errors, but that gives a rich signal to compensate for the physics discrepancy. Use another neural net to learn the delta. Basically, we "patch up" a traditional physics engine, so that the robot can experience almost the real world at scale in GPUs. The future is hybrid simulation: combine the power of classical sim engines refined over decades and the uncanny ability of modern NNs to capture a messy world.

Jim Fan

566,544 次观看 • 1 年前

NVIDIA just announced EgoScale 🤖🧠 NVIDIA Research has uncovered a log-linear scaling law for robot dexterity by pretraining VLA models on over 20,000 hours of egocentric human video This massive dataset is 20 times larger than previous efforts and proves that robot intelligence follows a predictable path: the more human data, the lower the loss The secret is a simple recipe combining large-scale human pretraining with a small amount of aligned human-robot mid-training to bridge the gap In testing, this method boosted the average success rate by 54% on a 22-DoF robotic hand compared to policies built without pretraining EgoScale also enables one-shot task adaptation and works across different hardware, suggesting that human motion is a universal motor prior for robots Website: Paper: Source: NVIDIA Research #Robot #Humanoid #Robotics #AI #EmbodiedAI #PhysicalAI #NVIDIA #EgoScale #GR00T
1:25

NVIDIA just announced EgoScale 🤖🧠 NVIDIA Research has uncovered a log-linear scaling law for robot dexterity by pretraining VLA models on over 20,000 hours of egocentric human video This massive dataset is 20 times larger than previous efforts and proves that robot intelligence follows a predictable path: the more human data, the lower the loss The secret is a simple recipe combining large-scale human pretraining with a small amount of aligned human-robot mid-training to bridge the gap In testing, this method boosted the average success rate by 54% on a 22-DoF robotic hand compared to policies built without pretraining EgoScale also enables one-shot task adaptation and works across different hardware, suggesting that human motion is a universal motor prior for robots Website: Paper: Source: NVIDIA Research #Robot #Humanoid #Robotics #AI #EmbodiedAI #PhysicalAI #NVIDIA #EgoScale #GR00T

RoboHub🤖

43,752 次观看 • 4 个月前

This is how ALOHA's "teleoperation" system works - a fancy word for "remote control". Training robots will be more and more like playing games in the physical world. A human operates a "joystick++" to perform tasks and collect data, or intervene if there's any safety concern. There's actually a learning curve to master the controller, much like practicing gaming skills. Teleoperation can be done in many different ways. ALOHA is an impressive custom-built system with very low cost. Here're a few alternatives: (1) Motion Capture (MoCap): apply the MoCap systems used for Hollywood movies to capture the fine-grained motions of hand joints. There would be no "embodiment gap" if the robot hand has 5 fingers. For instance, a demonstrator can wear a CyberGlove ( and manipulate the objects. CyberGlove will capture the motion signals & haptic feedback in real-time, which can be re-targeted onto the humanoid. (2) Wearing gloves & markers can be clumsy. An alternative way to do MoCap is through computer vision. DexPilot from NVIDIA enables marker-less and glove-free data collection. The human operator simply uses their bare hands to perform the tasks. 4 Intel RealSense depth cameras and 2 NVIDIA Titan XP GPUs (yeah, 2019 work) translate the pixels to precise motion signals for robot learning. (3) VR Headset: turn the training room into a VR game and "role play" the robot. This has the advantage of scalable remote data collection - annotators from around the world can contribute without coming onsite. VR demonstration technique appeared in research projects like the iGibson home robot simulator, an initiative that I participated in at Stanford: Behind-the-scene video by Litian Liang
0:52

This is how ALOHA's "teleoperation" system works - a fancy word for "remote control". Training robots will be more and more like playing games in the physical world. A human operates a "joystick++" to perform tasks and collect data, or intervene if there's any safety concern. There's actually a learning curve to master the controller, much like practicing gaming skills. Teleoperation can be done in many different ways. ALOHA is an impressive custom-built system with very low cost. Here're a few alternatives: (1) Motion Capture (MoCap): apply the MoCap systems used for Hollywood movies to capture the fine-grained motions of hand joints. There would be no "embodiment gap" if the robot hand has 5 fingers. For instance, a demonstrator can wear a CyberGlove ( and manipulate the objects. CyberGlove will capture the motion signals & haptic feedback in real-time, which can be re-targeted onto the humanoid. (2) Wearing gloves & markers can be clumsy. An alternative way to do MoCap is through computer vision. DexPilot from NVIDIA enables marker-less and glove-free data collection. The human operator simply uses their bare hands to perform the tasks. 4 Intel RealSense depth cameras and 2 NVIDIA Titan XP GPUs (yeah, 2019 work) translate the pixels to precise motion signals for robot learning. (3) VR Headset: turn the training room into a VR game and "role play" the robot. This has the advantage of scalable remote data collection - annotators from around the world can contribute without coming onsite. VR demonstration technique appeared in research projects like the iGibson home robot simulator, an initiative that I participated in at Stanford: Behind-the-scene video by Litian Liang

Jim Fan

124,588 次观看 • 2 年前

An interactive world model developed by NVIDIA in collaboration with academic partners. - DreamDojo turns egocentric human video data into physical intelligence. - Human data is more scalable than robotics data but lacks action labels. - To solve this, a dedicated action model extracts latent actions by identifying physics and motion deltas between frames. Training - A massive 44k hours of video data are used for pre-training. - Post-training on small-scale robot datasets maps human physics to specific robot embodiments. - An additional distillation stage converts the model into an autoregressive, few-step diffusion model, enabling real-time, action-controllable simulation. Primary Use Cases - Live Teleoperation: Controlling a robot inside a world simulation in real-time. - Model-based Planning: Previewing and curating the best actions for improved success. - Policy Evaluation: Testing robot policies in realistic, out-of-distribution scenarios. Everything that's open-sourced: weights, code, post-training dataset, eval set, and details to reproduce.
1:14

An interactive world model developed by NVIDIA in collaboration with academic partners. - DreamDojo turns egocentric human video data into physical intelligence. - Human data is more scalable than robotics data but lacks action labels. - To solve this, a dedicated action model extracts latent actions by identifying physics and motion deltas between frames. Training - A massive 44k hours of video data are used for pre-training. - Post-training on small-scale robot datasets maps human physics to specific robot embodiments. - An additional distillation stage converts the model into an autoregressive, few-step diffusion model, enabling real-time, action-controllable simulation. Primary Use Cases - Live Teleoperation: Controlling a robot inside a world simulation in real-time. - Model-based Planning: Previewing and curating the best actions for improved success. - Policy Evaluation: Testing robot policies in realistic, out-of-distribution scenarios. Everything that's open-sourced: weights, code, post-training dataset, eval set, and details to reproduce.

The Humanoid Hub

11,575 次观看 • 4 个月前

Synthetic data will provide the next trillion tokens to fuel our hungry models. I'm excited to announce MimicGen: massively scaling up data pipeline for robot learning! We multiply high-quality human data in simulation with digital twins. Using 50,000 training episodes across 18 tasks, multiple simulators, and even in the real-world! The idea is simple: 1. Humans tele-operate the robot to complete a task. It is extremely high-quality but also very slow and expensive. 2. We create a digital twin of the robot and the scene in high-fidelity, GPU-accelerated simulation. 3. We can now move objects around, replace with new assets, and even change the robot hand - basically augment the training data with procedural generation. 4. Export the successful episodes, and feed that to a neural network! You now have an near-infinite stream of data. One of the key reasons that robotics lags far behind other AI fields is the lack of data: you cannot scrape control signals from the internet. They simply don't exist in-the-wild. MimicGen shows the power of synthetic data and simulation to keep our scaling laws alive. I believe this principle apply beyond robotics. We are quickly exhausting the high-quality, real tokens from the web. Artificial intelligence from artificial data will be the way forward. We are big fans of the OSS community. As usual, we open-source everything, including the generated dataset! - Website: - Paper: - Dataset is hosted on HuggingFace (thanks AK!!): - Code: MimicGen is led by Ajay Mandlekar, deep dive in the thread:
0:46

Synthetic data will provide the next trillion tokens to fuel our hungry models. I'm excited to announce MimicGen: massively scaling up data pipeline for robot learning! We multiply high-quality human data in simulation with digital twins. Using 50,000 training episodes across 18 tasks, multiple simulators, and even in the real-world! The idea is simple: 1. Humans tele-operate the robot to complete a task. It is extremely high-quality but also very slow and expensive. 2. We create a digital twin of the robot and the scene in high-fidelity, GPU-accelerated simulation. 3. We can now move objects around, replace with new assets, and even change the robot hand - basically augment the training data with procedural generation. 4. Export the successful episodes, and feed that to a neural network! You now have an near-infinite stream of data. One of the key reasons that robotics lags far behind other AI fields is the lack of data: you cannot scrape control signals from the internet. They simply don't exist in-the-wild. MimicGen shows the power of synthetic data and simulation to keep our scaling laws alive. I believe this principle apply beyond robotics. We are quickly exhausting the high-quality, real tokens from the web. Artificial intelligence from artificial data will be the way forward. We are big fans of the OSS community. As usual, we open-source everything, including the generated dataset! - Website: - Paper: - Dataset is hosted on HuggingFace (thanks AK!!): - Code: MimicGen is led by Ajay Mandlekar, deep dive in the thread:

Jim Fan

332,199 次观看 • 2 年前

Excited to announce GR00T N1, the world’s first open foundation model for humanoid robots! We are on a mission to democratize Physical AI. The power of general robot brain, in the palm of your hand - with only 2B parameters, N1 learns from the most diverse physical action dataset ever compiled and punches above its weight: - Real humanoid teleoperation data. - Large-scale simulation data: we are open-sourcing 300K+ trajectories! - Neural trajectories: we apply SOTA video generation models to “hallucinate” new synthetic data that features accurate physics in pixels. Using Jensen’s words, “systematically infinite data”! - Latent actions: we develop novel algorithms to extract action tokens from in-the-wild human videos and neural generated videos. GR00T N1 is a single end-to-end neural net, from photons to actions: - Vision-Language Model (System 2) that interprets the physical world through vision and language instructions, enabling robots to reason about their environment and instructions, and plan the right actions. - Diffusion Transformer (System 1) that “renders” smooth and precise motor actions at 120 Hz, executing the latent plan made by System 2. We deploy N1 on GR1 robot, 1X Neo robot, and a large collection of simulation benchmarks. N1 achieves up to +30% boost in diverse manipulation tasks for household and industrial settings. While humanoid robots are the main focus of N1, our model also supports cross-embodiment. We finetune it to work on the $110 HuggingFace LeRobot SO100 robot arm! Open robot brain runs on open hardware. Sounds just right. Let’s solve robotics, together, one token at a time. Links to our Whitepaper, Github repo, HuggingFace model, and open dataset page in the thread: 🧵
2:21

Excited to announce GR00T N1, the world’s first open foundation model for humanoid robots! We are on a mission to democratize Physical AI. The power of general robot brain, in the palm of your hand - with only 2B parameters, N1 learns from the most diverse physical action dataset ever compiled and punches above its weight: - Real humanoid teleoperation data. - Large-scale simulation data: we are open-sourcing 300K+ trajectories! - Neural trajectories: we apply SOTA video generation models to “hallucinate” new synthetic data that features accurate physics in pixels. Using Jensen’s words, “systematically infinite data”! - Latent actions: we develop novel algorithms to extract action tokens from in-the-wild human videos and neural generated videos. GR00T N1 is a single end-to-end neural net, from photons to actions: - Vision-Language Model (System 2) that interprets the physical world through vision and language instructions, enabling robots to reason about their environment and instructions, and plan the right actions. - Diffusion Transformer (System 1) that “renders” smooth and precise motor actions at 120 Hz, executing the latent plan made by System 2. We deploy N1 on GR1 robot, 1X Neo robot, and a large collection of simulation benchmarks. N1 achieves up to +30% boost in diverse manipulation tasks for household and industrial settings. While humanoid robots are the main focus of N1, our model also supports cross-embodiment. We finetune it to work on the $110 HuggingFace LeRobot SO100 robot arm! Open robot brain runs on open hardware. Sounds just right. Let’s solve robotics, together, one token at a time. Links to our Whitepaper, Github repo, HuggingFace model, and open dataset page in the thread: 🧵

Jim Fan

465,968 次观看 • 1 年前

Robotics has a data scarcity problem - you simply can't scrape robot control data from webpages. Introducing GR00T-Mimic and GR00T-Gen: using both Graphics 1.0 & Graphics 2.0 to multiply your robot datasets by 1,000,000x. We trade compute for synthetic data, so we are not capped by the fundamental physical limit of 24 hrs/robot/day. Robotics is right in the thick of Moravec's paradox: things that are easy for humans turn out to be incredibly hard for machines. We are crushing the Moravec's paradox, one token at a time. > Graphics 1.0: Isaac simulators with manually written, GPU-accelerated physics and rendering equations. > Graphics 2.0: big neural nets (Cosmos) that repaint the pixels from sim textures to real, given an open-ended prompt. Robot data multiplier workflow: 1. GR00T-Teleop: use XR device like Apple Vision Pro to map human finger poses to humanoid hands. 2. GR00T-Mimic: given a human-collected task demonstration, we augment the actions in Isaac and filter out ones that fail the task. 3. GR00T-Gen: apply Graphics 1.0 and then Graphics 2.0 to produce tons of visual variations. Join us to build Physical AI together! Job application links in thread:
2:33

Robotics has a data scarcity problem - you simply can't scrape robot control data from webpages. Introducing GR00T-Mimic and GR00T-Gen: using both Graphics 1.0 & Graphics 2.0 to multiply your robot datasets by 1,000,000x. We trade compute for synthetic data, so we are not capped by the fundamental physical limit of 24 hrs/robot/day. Robotics is right in the thick of Moravec's paradox: things that are easy for humans turn out to be incredibly hard for machines. We are crushing the Moravec's paradox, one token at a time. > Graphics 1.0: Isaac simulators with manually written, GPU-accelerated physics and rendering equations. > Graphics 2.0: big neural nets (Cosmos) that repaint the pixels from sim textures to real, given an open-ended prompt. Robot data multiplier workflow: 1. GR00T-Teleop: use XR device like Apple Vision Pro to map human finger poses to humanoid hands. 2. GR00T-Mimic: given a human-collected task demonstration, we augment the actions in Isaac and filter out ones that fail the task. 3. GR00T-Gen: apply Graphics 1.0 and then Graphics 2.0 to produce tons of visual variations. Join us to build Physical AI together! Job application links in thread:

Jim Fan

163,670 次观看 • 1 年前

The World Model as NEO's Cognitive Core 1X has revealed a major AI development where the NEO humanoid can translate any natural language prompt into robotic action. It demonstrates this capability even for novel tasks, objects, and environments not found in its robot dataset. - the 1X World Model is trained on internet-scale human interaction videos and fine-tuned with robot data to ground its understanding in physics and in NEO's embodiment - from a simple voice or text prompt, the world model generates a visualization of future actions - a built-in inverse dynamics model then translates these into precise motor movements for NEO
1:02

The World Model as NEO's Cognitive Core 1X has revealed a major AI development where the NEO humanoid can translate any natural language prompt into robotic action. It demonstrates this capability even for novel tasks, objects, and environments not found in its robot dataset. - the 1X World Model is trained on internet-scale human interaction videos and fine-tuned with robot data to ground its understanding in physics and in NEO's embodiment - from a simple voice or text prompt, the world model generates a visualization of future actions - a built-in inverse dynamics model then translates these into precise motor movements for NEO

The Humanoid Hub

68,453 次观看 • 6 个月前

We trained a humanoid with 22-DoF dexterous hands to assemble model cars, operate syringes, sort poker cards, fold/roll shirts, all learned primarily from 20,000+ hours of egocentric human video with no robot in the loop. Humans are the most scalable embodiment on the planet. We discovered a near-perfect log-linear scaling law (R² = 0.998) between human video volume and action prediction loss, and this loss directly predicts real-robot success rate. Humanoid robots will be the end game, because they are the practical form factor with minimal embodiment gap from humans. Call it the Bitter Lesson of robot hardware: the kinematic similarity lets us simply retarget human finger motion onto dexterous robot hand joints. No learned embeddings, no fancy transfer algorithms needed. Relative wrist motion + retargeted 22-DoF finger actions serve as a unified action space that carries through from pre-training to robot execution. Our recipe is called "EgoScale": - Pre-train GR00T N1.5 on 20K hours of human video, mid-train with only 4 hours (!) of robot play data with Sharpa hands. 54% gains over training from scratch across 5 highly dexterous tasks. - Most surprising result: a *single* teleop demo is sufficient to learn a never-before-seen task. Our recipe enables extreme data efficiency. - Although we pre-train in 22-DoF hand joint space, the policy transfers to a Unitree G1 with 7-DoF tri-finger hands. 30%+ gains over training on G1 data alone. The scalable path to robot dexterity was never more robots. It was always us. Deep dives in thread:
1:17

We trained a humanoid with 22-DoF dexterous hands to assemble model cars, operate syringes, sort poker cards, fold/roll shirts, all learned primarily from 20,000+ hours of egocentric human video with no robot in the loop. Humans are the most scalable embodiment on the planet. We discovered a near-perfect log-linear scaling law (R² = 0.998) between human video volume and action prediction loss, and this loss directly predicts real-robot success rate. Humanoid robots will be the end game, because they are the practical form factor with minimal embodiment gap from humans. Call it the Bitter Lesson of robot hardware: the kinematic similarity lets us simply retarget human finger motion onto dexterous robot hand joints. No learned embeddings, no fancy transfer algorithms needed. Relative wrist motion + retargeted 22-DoF finger actions serve as a unified action space that carries through from pre-training to robot execution. Our recipe is called "EgoScale": - Pre-train GR00T N1.5 on 20K hours of human video, mid-train with only 4 hours (!) of robot play data with Sharpa hands. 54% gains over training from scratch across 5 highly dexterous tasks. - Most surprising result: a *single* teleop demo is sufficient to learn a never-before-seen task. Our recipe enables extreme data efficiency. - Although we pre-train in 22-DoF hand joint space, the policy transfers to a Unitree G1 with 7-DoF tri-finger hands. 30%+ gains over training on G1 data alone. The scalable path to robot dexterity was never more robots. It was always us. Deep dives in thread:

Jim Fan

292,967 次观看 • 4 个月前

Most RL locomotion examples let the actor (the policy network that runs on the real robot) observe two ground truths that are not directly measured by hardware: - linear velocity of the robot - projected gravity (i.e. orientation of the robot) The former can be inferred using a state estimator built using a small neural network trained to predict velocity, while the latter can be computed using Madgwick AHRS / Kalman filter. Alternatively, it kind of makes sense to let the actor network learn to extract whatever internal representation it needs directly from raw sensor data, instead of using hand-designed estimators. I removed base_lin_vel, similarly to Asimov's approach, as well as projected_gravity. Instead, I added the accelerometer data (which most RL examples do not seem to provide). I continue to give those ground truth variables to the critic as privileged info the actor can't see, which is known as an asymmetric actor-critic architecture. Advantages: 1. Should minimize the sim2real gap, as there are less external components whose results may differ between the sim and the hw 2. The actor can learn the interim representation that works better for the task, not necessarily those that we decided to infer for it 3. Less hand-tuned parameters At least in simulation this seems to work great. It might be luck, trivial or still plain wrong, but after 1500 iterations, the simulation reached the best run yet in terms of reward, lin/ang tracking, action std and more.
0:23

Most RL locomotion examples let the actor (the policy network that runs on the real robot) observe two ground truths that are not directly measured by hardware: - linear velocity of the robot - projected gravity (i.e. orientation of the robot) The former can be inferred using a state estimator built using a small neural network trained to predict velocity, while the latter can be computed using Madgwick AHRS / Kalman filter. Alternatively, it kind of makes sense to let the actor network learn to extract whatever internal representation it needs directly from raw sensor data, instead of using hand-designed estimators. I removed base_lin_vel, similarly to Asimov's approach, as well as projected_gravity. Instead, I added the accelerometer data (which most RL examples do not seem to provide). I continue to give those ground truth variables to the critic as privileged info the actor can't see, which is known as an asymmetric actor-critic architecture. Advantages: 1. Should minimize the sim2real gap, as there are less external components whose results may differ between the sim and the hw 2. The actor can learn the interim representation that works better for the task, not necessarily those that we decided to infer for it 3. Less hand-tuned parameters At least in simulation this seems to work great. It might be luck, trivial or still plain wrong, but after 1500 iterations, the simulation reached the best run yet in terms of reward, lin/ang tracking, action std and more.

David Bar

11,685 次观看 • 4 个月前

Can GPT-4 teach a robot hand to do pen spinning tricks better than you do? I'm excited to announce Eureka, an open-ended agent that designs reward functions for robot dexterity at super-human level. It’s like Voyager in the space of a physics simulator API! Eureka bridges the gap between high-level reasoning (coding) and low-level motor control. It is a “hybrid-gradient architecture”: a black box, inference-only LLM instructs a white box, learnable neural network. The outer loop runs GPT-4 to refine the reward function (gradient-free), while the inner loop runs reinforcement learning to train a robot controller (gradient-based). We are able to scale up Eureka thanks to IsaacGym, a GPU-accelerated physics simulator that speeds up reality by 1000x. On a benchmark suite of 29 tasks across 10 robots, Eureka rewards outperform expert human-written ones on 83% of the tasks by 52% improvement margin on average. We are surprised that Eureka is able to learn pen spinning tricks, which are very difficult even for CGI artists to animate frame by frame! Eureka also enables a new form of in-context RLHF, which is able to incorporate a human operator’s feedback in natural language to steer and align the reward functions. It can serve as a powerful co-pilot for robot engineers to design sophisticated motor behaviors. As usual, we open-source everything! Welcome you all to check out our video gallery and try the codebase today: Paper: Code: Deep dive with me: 🧵
0:24

Can GPT-4 teach a robot hand to do pen spinning tricks better than you do? I'm excited to announce Eureka, an open-ended agent that designs reward functions for robot dexterity at super-human level. It’s like Voyager in the space of a physics simulator API! Eureka bridges the gap between high-level reasoning (coding) and low-level motor control. It is a “hybrid-gradient architecture”: a black box, inference-only LLM instructs a white box, learnable neural network. The outer loop runs GPT-4 to refine the reward function (gradient-free), while the inner loop runs reinforcement learning to train a robot controller (gradient-based). We are able to scale up Eureka thanks to IsaacGym, a GPU-accelerated physics simulator that speeds up reality by 1000x. On a benchmark suite of 29 tasks across 10 robots, Eureka rewards outperform expert human-written ones on 83% of the tasks by 52% improvement margin on average. We are surprised that Eureka is able to learn pen spinning tricks, which are very difficult even for CGI artists to animate frame by frame! Eureka also enables a new form of in-context RLHF, which is able to incorporate a human operator’s feedback in natural language to steer and align the reward functions. It can serve as a powerful co-pilot for robot engineers to design sophisticated motor behaviors. As usual, we open-source everything! Welcome you all to check out our video gallery and try the codebase today: Paper: Code: Deep dive with me: 🧵

Jim Fan

2,673,990 次观看 • 2 年前

We trained a robot dog to balance and walk on top of a yoga ball purely in simulation, and then transfer zero-shot to the real world. No fine-tuning. Just works. I’m excited to announce DrEureka, an LLM agent that writes code to train robot skills in simulation, and writes more code to bridge the difficult simulation-reality gap. It fully automates the pipeline from new skill learning to real-world deployment. The Yoga ball task is particularly hard because it is not possible to accurately simulate the bouncy ball surface. Yet DrEureka has no trouble searching over a vast space of sim-to-real configurations, and enables the dog to steer the ball on various terrains, even walking sideways! Traditionally, the sim-to-real transfer is achieved by domain randomization, a tedious process that requires expert human roboticists to stare at every parameter and adjust by hand. Frontier LLMs like GPT-4 have tons of built-in physical intuition for friction, damping, stiffness, gravity, etc. We are (mildly) surprised to find that DrEureka can tune these parameters competently and explain its reasoning well. DrEureka builds on our prior work Eureka, the algorithm that teaches a 5-finger robot hand to do pen spinning. It takes one step further on our quest to automate the entire robot learning pipeline by an AI agent system. One model that outputs strings will supervise another model that outputs torque control. We open-source everything! Welcome you all to check out the paper, more videos, and try the codebase today: Code:
0:24

We trained a robot dog to balance and walk on top of a yoga ball purely in simulation, and then transfer zero-shot to the real world. No fine-tuning. Just works. I’m excited to announce DrEureka, an LLM agent that writes code to train robot skills in simulation, and writes more code to bridge the difficult simulation-reality gap. It fully automates the pipeline from new skill learning to real-world deployment. The Yoga ball task is particularly hard because it is not possible to accurately simulate the bouncy ball surface. Yet DrEureka has no trouble searching over a vast space of sim-to-real configurations, and enables the dog to steer the ball on various terrains, even walking sideways! Traditionally, the sim-to-real transfer is achieved by domain randomization, a tedious process that requires expert human roboticists to stare at every parameter and adjust by hand. Frontier LLMs like GPT-4 have tons of built-in physical intuition for friction, damping, stiffness, gravity, etc. We are (mildly) surprised to find that DrEureka can tune these parameters competently and explain its reasoning well. DrEureka builds on our prior work Eureka, the algorithm that teaches a 5-finger robot hand to do pen spinning. It takes one step further on our quest to automate the entire robot learning pipeline by an AI agent system. One model that outputs strings will supervise another model that outputs torque control. We open-source everything! Welcome you all to check out the paper, more videos, and try the codebase today: Code:

Jim Fan

908,690 次观看 • 2 年前

Experiments in progress. The one on the right has been learning for ~3 hours, the one in the middle for ~1 hour, and the one on the left just started a few minutes ago. The initial motivation for making the physical Atari was just to commit ourselves to a subset of algorithms that can make progress in this setup. This commitment rules out algorithms that require billions of samples to learn (or worse, require multiple environments running in parallel). Atari games are simple enough that we should be able to show learning on them in a short amount of time with no prior knowledge. Since then, I've realized that this setup is also a good way to compare different paradigms in robotics in a principled way. These paradigms are sim2real, learning from tele-operated data, and learning directly on the robots. So far, I have observed that getting sim2real to work reliably is hard. It requires tweaks that don't scale. Policies that can play perfectly in simulation fall apart because of latencies and the messiness of the real world. These aspects could be modeled to improve the simulation, but not without sinking significant human engineering hours. I have higher hopes for learning from tele-operated data, but that requires a human to learn the task first. These experiments are on my to-do list. I have to learn to play some of the games well through the robot. I’m half-decent at playing Pong and Ms Pacman now. Learning directly on robots is looking like the most promising approach. This approach takes away pesky distribution shifts and makes it possible to have algorithms that continually improve with more data and time without any human intervention. It feels great to let experiments run overnight and wake up to find improved policies. With learning on robots, I should, in principle, be able to go on a long vacation and come back to find better policies for complex tasks beyond Atari games. Whether that is possible with current learning algorithms is a different question.
1:05

Experiments in progress. The one on the right has been learning for ~3 hours, the one in the middle for ~1 hour, and the one on the left just started a few minutes ago. The initial motivation for making the physical Atari was just to commit ourselves to a subset of algorithms that can make progress in this setup. This commitment rules out algorithms that require billions of samples to learn (or worse, require multiple environments running in parallel). Atari games are simple enough that we should be able to show learning on them in a short amount of time with no prior knowledge. Since then, I've realized that this setup is also a good way to compare different paradigms in robotics in a principled way. These paradigms are sim2real, learning from tele-operated data, and learning directly on the robots. So far, I have observed that getting sim2real to work reliably is hard. It requires tweaks that don't scale. Policies that can play perfectly in simulation fall apart because of latencies and the messiness of the real world. These aspects could be modeled to improve the simulation, but not without sinking significant human engineering hours. I have higher hopes for learning from tele-operated data, but that requires a human to learn the task first. These experiments are on my to-do list. I have to learn to play some of the games well through the robot. I’m half-decent at playing Pong and Ms Pacman now. Learning directly on robots is looking like the most promising approach. This approach takes away pesky distribution shifts and makes it possible to have algorithms that continually improve with more data and time without any human intervention. It feels great to let experiments run overnight and wake up to find improved policies. With learning on robots, I should, in principle, be able to go on a long vacation and come back to find better policies for complex tasks beyond Atari games. Whether that is possible with current learning algorithms is a different question.

Khurram Javed

52,110 次观看 • 7 个月前

Everything Elon said about Optimus at the All-In Summit today: • We’re finalizing the design of Optimus v3. That release is going to be a very remarkable robot. It will have manual dexterity comparable to a human, meaning a very complex hand, an AI mind that can navigate and comprehend reality, and will be made in very high volume. • Other robotics companies are missing those three very hard things. • I spend more mental cycles on Optimus than any other single thing. Solving real-world AI, all of the electrical-mechanical issues, the supply chain, and production challenges. • There is no supply chain for humanoid robots, so it has to be created from scratch, which requires a lot of vertical integration. None of the actuators in Optimus are available from an existing supply chain. • I think if successful, Optimus would be the biggest product ever. • The marginal cost of production, once we hit a million units per year, will probably be around $20,000. It depends on how much we spend on the AI chip in the robot, and we’ll need to achieve a lot of efficiencies in the actuators—26 actuators per arm (26 motors, gearboxes, and power electronics). The AI chip might cost $5,000 or $6,000, maybe more. At 1 million units a year, production cost will be $20,000, maybe $25,000. Price will be a function of demand. • Human hands have evolved to be incredibly sophisticated machines. Hands are a very first instrument. You can swing a baseball bat, thread a needle, play a piano or violin, and assemble a car. Hands are incredibly versatile instruments. Most of the muscles of the hands are actually in the forearm, and the hand is almost like a puppet. Human tendon evolution is incredibly good. The human hand has 27 or 28 degrees of freedom, depending on how you count it; it’s amazing. • In order to create a robot that can be a generalized humanoid, you must solve the “hands problem.” • Even though there are 10,000 to 20,000 electric motors out there, we couldn’t buy the actuators for any amount of money. We had to design every electric motor, gearbox, and controlling electronics from scratch, from first principles of physics. • Optimus is harder than developing any previous Tesla product, but not harder than Starship. • Right now, we’re struggling with the final design of the hardware, primarily the hand. The hands and forearm are the majority of the engineering difficulty of the entire robot. • If you want to do all the things that a human can do, it turns out you need a humanoid robot. If you want to do a subset, that’s much easier. Humans evolved to the shape and capability that we have for a good reason. There is value to having four fingers and a thumb; even the pinky is quite useful. Toes are much more of a question mark. • The AI5 inference chip will be 40 times better than AI4 by some measures. We know the limiting factors of the chip because the AI software and hardware teams work so closely. Effectively, the Tesla AI hardware and software teams are co-designing the chip. • The Softmax function on AI4 takes 40 steps in emulation mode, which will take only a few steps in AI5 natively. AI5 will easily handle mixed precision. • In terms of nominal raw compute, the AI5 inference chip has 8 times more compute, 9 times more memory, and 5 times more memory bandwidth compared to AI4. Because we’re addressing some core limitations and optimizations at the silicon level, we’re able to realize 40x improvements.
12:26

Everything Elon said about Optimus at the All-In Summit today: • We’re finalizing the design of Optimus v3. That release is going to be a very remarkable robot. It will have manual dexterity comparable to a human, meaning a very complex hand, an AI mind that can navigate and comprehend reality, and will be made in very high volume. • Other robotics companies are missing those three very hard things. • I spend more mental cycles on Optimus than any other single thing. Solving real-world AI, all of the electrical-mechanical issues, the supply chain, and production challenges. • There is no supply chain for humanoid robots, so it has to be created from scratch, which requires a lot of vertical integration. None of the actuators in Optimus are available from an existing supply chain. • I think if successful, Optimus would be the biggest product ever. • The marginal cost of production, once we hit a million units per year, will probably be around $20,000. It depends on how much we spend on the AI chip in the robot, and we’ll need to achieve a lot of efficiencies in the actuators—26 actuators per arm (26 motors, gearboxes, and power electronics). The AI chip might cost $5,000 or $6,000, maybe more. At 1 million units a year, production cost will be $20,000, maybe $25,000. Price will be a function of demand. • Human hands have evolved to be incredibly sophisticated machines. Hands are a very first instrument. You can swing a baseball bat, thread a needle, play a piano or violin, and assemble a car. Hands are incredibly versatile instruments. Most of the muscles of the hands are actually in the forearm, and the hand is almost like a puppet. Human tendon evolution is incredibly good. The human hand has 27 or 28 degrees of freedom, depending on how you count it; it’s amazing. • In order to create a robot that can be a generalized humanoid, you must solve the “hands problem.” • Even though there are 10,000 to 20,000 electric motors out there, we couldn’t buy the actuators for any amount of money. We had to design every electric motor, gearbox, and controlling electronics from scratch, from first principles of physics. • Optimus is harder than developing any previous Tesla product, but not harder than Starship. • Right now, we’re struggling with the final design of the hardware, primarily the hand. The hands and forearm are the majority of the engineering difficulty of the entire robot. • If you want to do all the things that a human can do, it turns out you need a humanoid robot. If you want to do a subset, that’s much easier. Humans evolved to the shape and capability that we have for a good reason. There is value to having four fingers and a thumb; even the pinky is quite useful. Toes are much more of a question mark. • The AI5 inference chip will be 40 times better than AI4 by some measures. We know the limiting factors of the chip because the AI software and hardware teams work so closely. Effectively, the Tesla AI hardware and software teams are co-designing the chip. • The Softmax function on AI4 takes 40 steps in emulation mode, which will take only a few steps in AI5 natively. AI5 will easily handle mixed precision. • In terms of nominal raw compute, the AI5 inference chip has 8 times more compute, 9 times more memory, and 5 times more memory bandwidth compared to AI4. Because we’re addressing some core limitations and optimizations at the silicon level, we’re able to realize 40x improvements.

The Humanoid Hub

238,630 次观看 • 10 个月前