Загрузка видео...

Не удалось загрузить видео

Возникла проблема при загрузке этого видео. Это может быть связано с временными проблемами сети или видео может быть недоступно.

На главную

Having humans annotate data to pre-train robots is expensive and time-consuming! Introducing SPRINT: A pre-training approach using LLMs and offline RL to equip robots w/ many language-annotated skills while minimizing human annotation effort! URL: 🧵👇

Jesse Zhang @ ICML 2026 ✈️

1,711 subscribers

24,665 просмотров • 3 лет назад •via X (Twitter)

Anya Rossi• Live Now

Private livecam show

Комментарии: 7

Фото профиля Jesse Zhang

Jesse Zhang3 лет назад

Labeling demonstrations with natural language instructions in hindsight is standard, but it is tedious and expensive to scale. We propose automatically (1) **relabeling** language instructions and (2) **chaining** trajectories together to generate more training data.

Фото профиля Jesse Zhang

Jesse Zhang3 лет назад

(1) Relabeling: If we have two skills, "Put mug in coffee machine" and "Press brew button," we could call this "Make Coffee." In SPRINT, we do this relabeling **automatically** by prompting an LLM to summarize nearby instructions. This gives us 2-2.5X more pre-training data!

Фото профиля Jesse Zhang

Jesse Zhang3 лет назад

(2) Chaining: Offline RL can "stitch" trajectories to learn new behaviors. We carefully relabel rewards with offline RL and modified language instructions to allow stitching even with language conditioning!

Фото профиля Jesse Zhang

Jesse Zhang3 лет назад

Results Overall, this allows us to achieve 2-8X better zero-shot long horizon task execution in ALFRED, a realistic household simulator, and on a real robot setup! SPRINT agents also fine-tune more efficiently to new tasks in unseen environments! ALFRED results:

Фото профиля Jesse Zhang

Jesse Zhang3 лет назад

Real Robot Results With offline fine-tuning, SPRINT achieves superior performance on new, long-horizon manipulation tasks in previously unseen environments!

Фото профиля Jesse Zhang

Jesse Zhang3 лет назад

For more details about SPRINT and experiment results, please see our paper or website. Paper: Website: Work done in collaboration with @KarlPertsch, @JiahuiZhang_32, @JosephLim_AI. @JiahuiZhang_32 is applying for PhD this year!

Фото профиля OliviaLi

OliviaLi2 лет назад

It sounds great, so that humans don't have to complete such a large amount of work every day, just let the robot do it

Похожие видео

How to teach robots to perform long-horizon tasks efficiently and robustly🦾? Introducing MimicPlay - an imitation learning algorithm that uses "cheap human play data". Our approach unlocks both real-time planning through raw perception and strong robustness to disturbances!🧵👇

How to teach robots to perform long-horizon tasks efficiently and robustly🦾? Introducing MimicPlay - an imitation learning algorithm that uses "cheap human play data". Our approach unlocks both real-time planning through raw perception and strong robustness to disturbances!🧵👇

Chen Wang

288,642 просмотров • 3 лет назад

Wouldn't it be great if we could train robots without any teleoperation! In our latest paper, we train robots to mimic a human video of the task by simply matching the object features using RL. We only need one video and under an hour of robot training.

Wouldn't it be great if we could train robots without any teleoperation! In our latest paper, we train robots to mimic a human video of the task by simply matching the object features using RL. We only need one video and under an hour of robot training.

Lerrel Pinto

46,221 просмотров • 1 год назад

HDMI (HumanoiD iMitation for Interaction) is a framework enabling humanoid robots to learn whole-body object interaction skills from monocular RGB human videos. It extracts and retargets human poses and object trajectories using GVHMR and LocoMujoco, building reference datasets with contact annotations. The data is used to train an RL policy via robot-object co-tracking. HDMI achieved 67 consecutive door traversals.

HDMI (HumanoiD iMitation for Interaction) is a framework enabling humanoid robots to learn whole-body object interaction skills from monocular RGB human videos. It extracts and retargets human poses and object trajectories using GVHMR and LocoMujoco, building reference datasets with contact annotations. The data is used to train an RL policy via robot-object co-tracking. HDMI achieved 67 consecutive door traversals.

The Humanoid Hub

17,395 просмотров • 9 месяцев назад

Amazon is training humanoids to move boxes. Makes sense! OmniRetarget is a data generation engine that enables complex loco-manipulation for humanoids. It uses offline retargeting from human MoCap datasets and augments data from single demos to produce 8 hours of trajectories that train RL policies in simulation, transferred zero-shot to real robots. Amazon FAR is the primary affiliation for the paper:

Amazon is training humanoids to move boxes. Makes sense! OmniRetarget is a data generation engine that enables complex loco-manipulation for humanoids. It uses offline retargeting from human MoCap datasets and augments data from single demos to produce 8 hours of trajectories that train RL policies in simulation, transferred zero-shot to real robots. Amazon FAR is the primary affiliation for the paper:

The Humanoid Hub

106,447 просмотров • 9 месяцев назад

One of the biggest bottlenecks in deploying visual AI and computer vision is annotation, which can be both costly and time-consuming. Today, we’re introducing Verified Auto Labeling, a new approach to AI-assisted annotation that achieves up to 95% of human-level performance while cutting labeling costs by up to 100,000x and time by 5,000x. Read the full paper:

One of the biggest bottlenecks in deploying visual AI and computer vision is annotation, which can be both costly and time-consuming. Today, we’re introducing Verified Auto Labeling, a new approach to AI-assisted annotation that achieves up to 95% of human-level performance while cutting labeling costs by up to 100,000x and time by 5,000x. Read the full paper:

Voxel51

12,708 просмотров • 1 год назад

Our course recommendation of the day is “Post-training of LLMs, ” where you’ll learn how to customize pre-trained language models using Supervised Fine-Tuning (SFT), Direct Preference Optimization (DPO), and Online Reinforcement Learning (RL). You'll learn when to use each method, how to curate training data, and implement them in code to shape model behavior effectively. Enroll here:

Our course recommendation of the day is “Post-training of LLMs, ” where you’ll learn how to customize pre-trained language models using Supervised Fine-Tuning (SFT), Direct Preference Optimization (DPO), and Online Reinforcement Learning (RL). You'll learn when to use each method, how to curate training data, and implement them in code to shape model behavior effectively. Enroll here:

DeepLearning.AI

29,369 просмотров • 9 месяцев назад

Brain-controlled exoskeletons to train humanoid robots! 🧠 Fourier just presented human tele-operators using brain control interfaces and exoskeletal arms to train humanoid robots on home tasks. The brain control interface is the interesting part. Instead of using a controller or joystick to teleoperate, the operator's movements and intentions are captured more naturally through the exoskeleton and BCI. This means the demonstrations are more fluid, more human-like, and better suited for training robots to perform delicate home tasks. Multiple tele-operators are simultaneously generating training data across multiple robots. This is how you build the dataset needed for eventual full autonomy, without waiting years for it to arrive. This might be the bridge between "robots that work in controlled environments" and "robots that work in homes." Not full autonomy right away, but trusted human intelligence operating through a robot body, getting better with every task completed. ~~ ♻️ Join the weekly robotics newsletter, and never miss any news →

Brain-controlled exoskeletons to train humanoid robots! 🧠 Fourier just presented human tele-operators using brain control interfaces and exoskeletal arms to train humanoid robots on home tasks. The brain control interface is the interesting part. Instead of using a controller or joystick to teleoperate, the operator's movements and intentions are captured more naturally through the exoskeleton and BCI. This means the demonstrations are more fluid, more human-like, and better suited for training robots to perform delicate home tasks. Multiple tele-operators are simultaneously generating training data across multiple robots. This is how you build the dataset needed for eventual full autonomy, without waiting years for it to arrive. This might be the bridge between "robots that work in controlled environments" and "robots that work in homes." Not full autonomy right away, but trusted human intelligence operating through a robot body, getting better with every task completed. ~~ ♻️ Join the weekly robotics newsletter, and never miss any news →

Lukas Ziegler

20,898 просмотров • 5 месяцев назад

We discovered an emergent property of VLAs like π0/π0.5/π0.6: as we scale up pre-training, the model learns to align human videos and robot data! This gives us a simple way to leverage human videos. Once π0.5 knows how to control robots, it can naturally learn from human video.

We discovered an emergent property of VLAs like π0/π0.5/π0.6: as we scale up pre-training, the model learns to align human videos and robot data! This gives us a simple way to leverage human videos. Once π0.5 knows how to control robots, it can naturally learn from human video.

Physical Intelligence

1,182,874 просмотров • 6 месяцев назад

Introducing LDA, a latent world action foundation model that, for the first time, unifies the utilization of heterogeneous embodied data across simulation and reality, humans and robots, and varying levels of action quality and annotation. By breaking long-standing data silos in embodied intelligence, LDA enables the field, much like GPT did for language, to benefit continuously from scaling data, marking the transition into a new era of scalable learning. #Galbot #Robotics #Innovation #AI #Technology #Humanoid #WorldModel

Introducing LDA, a latent world action foundation model that, for the first time, unifies the utilization of heterogeneous embodied data across simulation and reality, humans and robots, and varying levels of action quality and annotation. By breaking long-standing data silos in embodied intelligence, LDA enables the field, much like GPT did for language, to benefit continuously from scaling data, marking the transition into a new era of scalable learning. #Galbot #Robotics #Innovation #AI #Technology #Humanoid #WorldModel

Galbot

38,131 просмотров • 2 месяцев назад

NEWS: DeepMind teaches robots to play soccer. In a paper released today Google's DeepMind details how they used Deep Reinforcement Learning (Deep RL) to train low-cost, miniature humanoid robots in dynamic environments, allowing them to play a simplified one-versus-one (1v1) soccer game. The robots, equipped with 20 actuated joints, were initially trained in simulation using the MuJoCo physics engine. Through this training, they learned robust and dynamic movement skills like rapid fall recovery, walking, turning, and kicking. The robots seamlessly transitioned between these skills, even surpassing expectations, and developed a basic strategic understanding of the game. During matches, the trained robots demonstrated agile skills such as turning, kicking moving balls, and dynamic defensive blocking. They quickly combined these skills, showcasing their adaptability and outperforming scripted baselines. The robots walked faster, got up quicker, and kicked faster than their counterparts. Individual skills were first trained in isolation within the simulation environment, and then composed in a self-play setting. The robots successfully transferred these skills to real-world scenarios. Full paper can be found here:

NEWS: DeepMind teaches robots to play soccer. In a paper released today Google's DeepMind details how they used Deep Reinforcement Learning (Deep RL) to train low-cost, miniature humanoid robots in dynamic environments, allowing them to play a simplified one-versus-one (1v1) soccer game. The robots, equipped with 20 actuated joints, were initially trained in simulation using the MuJoCo physics engine. Through this training, they learned robust and dynamic movement skills like rapid fall recovery, walking, turning, and kicking. The robots seamlessly transitioned between these skills, even surpassing expectations, and developed a basic strategic understanding of the game. During matches, the trained robots demonstrated agile skills such as turning, kicking moving balls, and dynamic defensive blocking. They quickly combined these skills, showcasing their adaptability and outperforming scripted baselines. The robots walked faster, got up quicker, and kicked faster than their counterparts. Individual skills were first trained in isolation within the simulation environment, and then composed in a self-play setting. The robots successfully transferred these skills to real-world scenarios. Full paper can be found here:

Dave Lee

372,039 просмотров • 3 лет назад

TWIST: a real-time teleoperation system for humanoid robots to mimic whole-body motions. Reference motion data is generated by retargeting human motion-capture data to the robot. Then, the controller is trained in simulation using reinforcement learning and behavior cloning.

TWIST: a real-time teleoperation system for humanoid robots to mimic whole-body motions. Reference motion data is generated by retargeting human motion-capture data to the robot. Then, the controller is trained in simulation using reinforcement learning and behavior cloning.

The Humanoid Hub

46,715 просмотров • 1 год назад

Introducing Yell At Your Robot (YAY Robot!) 🗣️- a fun collaboration b/w Stanford University and UC Berkeley 🤖 We enable robots to improve on-the-fly from language corrections: robots rapidly adapt in real-time and continuously improve from human verbal feedback. YAY Robot enables long-horizon, dexterous manipulation tasks like preparing trail-mix, packing a ziploc bag, and cleaning dishes:

Introducing Yell At Your Robot (YAY Robot!) 🗣️- a fun collaboration b/w Stanford University and UC Berkeley 🤖 We enable robots to improve on-the-fly from language corrections: robots rapidly adapt in real-time and continuously improve from human verbal feedback. YAY Robot enables long-horizon, dexterous manipulation tasks like preparing trail-mix, packing a ziploc bag, and cleaning dishes:

Lucy Shi

122,827 просмотров • 2 лет назад

New Course: Post-training of LLMs Learn to post-train and customize an LLM in this short course, taught by Banghua Zhu, Assistant Professor at the University of Washington University of Washington, and co-founder of @NexusflowX. Training an LLM to follow instructions or answer questions has two key stages: pre-training and post-training. In pre-training, it learns to predict the next word or token from large amounts of unlabeled text. In post-training, it learns useful behaviors such as following instructions, tool use, and reasoning. Post-training transforms a general-purpose token predictor—trained on trillions of unlabeled text tokens—into an assistant that follows instructions and performs specific tasks. Because it is much cheaper than pre-training, it is practical for many more teams to incorporate post-training methods into their workflows than pre-training. In this course, you’ll learn three common post-training methods—Supervised Fine-Tuning (SFT), Direct Preference Optimization (DPO), and Online Reinforcement Learning (RL)—and how to use each one effectively. With SFT, you train the model on pairs of input and ideal output responses. With DPO, you provide both a preferred (chosen) and a less preferred (rejected) response and train the model to favor the preferred output. With RL, the model generates an output, receives a reward score based on human or automated feedback, and updates the model to improve performance. You’ll learn the basic concepts, common use cases, and principles for curating high-quality data for effective training. Through hands-on labs, you’ll download a pre-trained model from Hugging Face and post-train it using SFT, DPO, and RL to see how each technique shapes model behavior. In detail, you’ll: - Understand what post-training is, when to use it, and how it differs from pre-training. - Build an SFT pipeline to turn a base model into an instruct model. - Explore how DPO reshapes behavior by minimizing contrastive loss—penalizing poor responses and reinforcing preferred ones. - Implement a DPO pipeline to change the identity of a chat assistant. - Learn online RL methods such as Proximal Policy Optimization (PPO) and Group Relative Policy Optimization (GRPO), and how to design reward functions. - Train a model with GRPO to improve its math capabilities using a verifiable reward. Post-training is one of the most rapidly developing areas of LLM training. Whether you’re building a high-accuracy context-specific assistant, fine-tuning a model's tone, or improving task-specific accuracy, this course will give you experience with the most important techniques shaping how LLMs are post-trained today. Please sign up here:

New Course: Post-training of LLMs Learn to post-train and customize an LLM in this short course, taught by Banghua Zhu, Assistant Professor at the University of Washington University of Washington, and co-founder of @NexusflowX. Training an LLM to follow instructions or answer questions has two key stages: pre-training and post-training. In pre-training, it learns to predict the next word or token from large amounts of unlabeled text. In post-training, it learns useful behaviors such as following instructions, tool use, and reasoning. Post-training transforms a general-purpose token predictor—trained on trillions of unlabeled text tokens—into an assistant that follows instructions and performs specific tasks. Because it is much cheaper than pre-training, it is practical for many more teams to incorporate post-training methods into their workflows than pre-training. In this course, you’ll learn three common post-training methods—Supervised Fine-Tuning (SFT), Direct Preference Optimization (DPO), and Online Reinforcement Learning (RL)—and how to use each one effectively. With SFT, you train the model on pairs of input and ideal output responses. With DPO, you provide both a preferred (chosen) and a less preferred (rejected) response and train the model to favor the preferred output. With RL, the model generates an output, receives a reward score based on human or automated feedback, and updates the model to improve performance. You’ll learn the basic concepts, common use cases, and principles for curating high-quality data for effective training. Through hands-on labs, you’ll download a pre-trained model from Hugging Face and post-train it using SFT, DPO, and RL to see how each technique shapes model behavior. In detail, you’ll: - Understand what post-training is, when to use it, and how it differs from pre-training. - Build an SFT pipeline to turn a base model into an instruct model. - Explore how DPO reshapes behavior by minimizing contrastive loss—penalizing poor responses and reinforcing preferred ones. - Implement a DPO pipeline to change the identity of a chat assistant. - Learn online RL methods such as Proximal Policy Optimization (PPO) and Group Relative Policy Optimization (GRPO), and how to design reward functions. - Train a model with GRPO to improve its math capabilities using a verifiable reward. Post-training is one of the most rapidly developing areas of LLM training. Whether you’re building a high-accuracy context-specific assistant, fine-tuning a model's tone, or improving task-specific accuracy, this course will give you experience with the most important techniques shaping how LLMs are post-trained today. Please sign up here:

Andrew Ng

125,146 просмотров • 1 год назад

🤖 Introducing H2O (Human2HumanOid): - 🧠 An RL-based human-to-humanoid real-time whole-body teleoperation framework - 💃 Scalable retargeting and training using large human motion dataset - 🎥 With just an RGB camera, everyone can teleoperate a full-sized humanoid to perform actions like pick and place, walking, kicking, boxing, etc - 💡Unleash the potential of humanoids with human cognitive skills and adaptability 🔗:

🤖 Introducing H2O (Human2HumanOid): - 🧠 An RL-based human-to-humanoid real-time whole-body teleoperation framework - 💃 Scalable retargeting and training using large human motion dataset - 🎥 With just an RGB camera, everyone can teleoperate a full-sized humanoid to perform actions like pick and place, walking, kicking, boxing, etc - 💡Unleash the potential of humanoids with human cognitive skills and adaptability 🔗:

Tairan He

96,738 просмотров • 2 лет назад

Robots has a data problem, in that robotics data is rare. While human video is quite common, it’s not usually directly usable for robots for a variety of reasons, most significantly that it’s missing explicit, accurate robot actions. Instead, Jeremy Collins proposes that we predict keypoint trajectories — basically, how any given point in an object will move as a robot performs a task. This lets us use action-free human video to train robot skills. Learn more by watching Episode #37 of RoboPapers with Michael Cho - Rbt/Acc and Chris Paxton now.

Robots has a data problem, in that robotics data is rare. While human video is quite common, it’s not usually directly usable for robots for a variety of reasons, most significantly that it’s missing explicit, accurate robot actions. Instead, Jeremy Collins proposes that we predict keypoint trajectories — basically, how any given point in an object will move as a robot performs a task. This lets us use action-free human video to train robot skills. Learn more by watching Episode #37 of RoboPapers with Michael Cho - Rbt/Acc and Chris Paxton now.

RoboPapers

89,011 просмотров • 8 месяцев назад

Still following your human intuition to mix corpora from different sources for language model pre-training 🧠? Everyone says that data mixture has a big impact on model performance, but how - and why🕵️? Did you know that web corpora are actually highly impactful for downstream tasks 🏆? Let's check out our preprint "RegMix: Data Mixture as Regression for Language Model Pre-training" 📄 🔬In this paper, we've proposed an automatic data mixture method RegMix that achieves a 6.3% improvement over human selection on the widely used HellaSwag benchmark - and it only needs a 2% extra training FLOPs! 📈 Details in the thread 🧵

Still following your human intuition to mix corpora from different sources for language model pre-training 🧠? Everyone says that data mixture has a big impact on model performance, but how - and why🕵️? Did you know that web corpora are actually highly impactful for downstream tasks 🏆? Let's check out our preprint "RegMix: Data Mixture as Regression for Language Model Pre-training" 📄 🔬In this paper, we've proposed an automatic data mixture method RegMix that achieves a 6.3% improvement over human selection on the widely used HellaSwag benchmark - and it only needs a 2% extra training FLOPs! 📈 Details in the thread 🧵

Qian Liu

54,961 просмотров • 2 лет назад

In case you missed it, we recently launched "Post-training of LLMs," a short course where you'll: ✅ Understand when and why to use post-training methods like Supervised Fine-Tuning (SFT), Direct Preference Optimization (DPO), and Online Reinforcement Learning. ✅ Learn the concepts underlying the three post-training methods of SFT, DPO, and Online RL, their common use-cases, and how to curate high-quality data to effectively train a model using each method. ✅ Download a pre-trained model and implement post-training pipelines to turn a base model into an instruct model, change the identity of a chat assistant, and improve a model’s math capabilities. Learn more and enroll for free:

In case you missed it, we recently launched "Post-training of LLMs," a short course where you'll: ✅ Understand when and why to use post-training methods like Supervised Fine-Tuning (SFT), Direct Preference Optimization (DPO), and Online Reinforcement Learning. ✅ Learn the concepts underlying the three post-training methods of SFT, DPO, and Online RL, their common use-cases, and how to curate high-quality data to effectively train a model using each method. ✅ Download a pre-trained model and implement post-training pipelines to turn a base model into an instruct model, change the identity of a chat assistant, and improve a model’s math capabilities. Learn more and enroll for free:

DeepLearning.AI

16,771 просмотров • 1 год назад

New short course on Fine-tuning LLMs! Many developers are moving beyond only prompting, to also fine-tuning LLMs - that is, taking a pre-trained model and training it further on your own data, which can deliver superior results inexpensively. In this course, Sharon Zhou ✈️ ICML, CEO of Lamini (disclosure: I’m a minor shareholder) shows you how to recognize when fine-tuning can be help, and how to train an open-source LLM on your own data. I hope you enjoy the course!

New short course on Fine-tuning LLMs! Many developers are moving beyond only prompting, to also fine-tuning LLMs - that is, taking a pre-trained model and training it further on your own data, which can deliver superior results inexpensively. In this course, Sharon Zhou ✈️ ICML, CEO of Lamini (disclosure: I’m a minor shareholder) shows you how to recognize when fine-tuning can be help, and how to train an open-source LLM on your own data. I hope you enjoy the course!

Andrew Ng

502,810 просмотров • 2 лет назад

MotionGPT: Human Motion as a Foreign Language paper page: Though the advancement of pre-trained large language models unfolds, the exploration of building a unified model for language and other multi-modal data, such as motion, remains challenging and untouched so far. Fortunately, human motion displays a semantic coupling akin to human language, often perceived as a form of body language. By fusing language data with large-scale motion models, motion-language pre-training that can enhance the performance of motion-related tasks becomes feasible. Driven by this insight, we propose MotionGPT, a unified, versatile, and user-friendly motion-language model to handle multiple motion-relevant tasks. Specifically, we employ the discrete vector quantization for human motion and transfer 3D motion into motion tokens, similar to the generation process of word tokens. Building upon this "motion vocabulary", we perform language modeling on both motion and text in a unified manner, treating human motion as a specific language. Moreover, inspired by prompt learning, we pre-train MotionGPT with a mixture of motion-language data and fine-tune it on prompt-based question-and-answer tasks. Extensive experiments demonstrate that MotionGPT achieves state-of-the-art performances on multiple motion tasks including text-driven motion generation, motion captioning, motion prediction, and motion in-between.

MotionGPT: Human Motion as a Foreign Language paper page: Though the advancement of pre-trained large language models unfolds, the exploration of building a unified model for language and other multi-modal data, such as motion, remains challenging and untouched so far. Fortunately, human motion displays a semantic coupling akin to human language, often perceived as a form of body language. By fusing language data with large-scale motion models, motion-language pre-training that can enhance the performance of motion-related tasks becomes feasible. Driven by this insight, we propose MotionGPT, a unified, versatile, and user-friendly motion-language model to handle multiple motion-relevant tasks. Specifically, we employ the discrete vector quantization for human motion and transfer 3D motion into motion tokens, similar to the generation process of word tokens. Building upon this "motion vocabulary", we perform language modeling on both motion and text in a unified manner, treating human motion as a specific language. Moreover, inspired by prompt learning, we pre-train MotionGPT with a mixture of motion-language data and fine-tune it on prompt-based question-and-answer tasks. Extensive experiments demonstrate that MotionGPT achieves state-of-the-art performances on multiple motion tasks including text-driven motion generation, motion captioning, motion prediction, and motion in-between.

AK

125,319 просмотров • 3 лет назад

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:

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 просмотров • 4 месяцев назад