Video wird geladen...

Video konnte nicht geladen werden

Zur Startseite

Before a robot can perfect assembly, it needs to learn to play. The team behind SimToolReal Kushal Tyler Lum Jeannette Bohg Prof Karen J Liu published another cool paper! Play2Perfect pretrains on diverse, task-agnostic play (grasp, reorient, reach, etc), then finetunes on sparse-reward assembly. Result: 33× sample efficiency vs....

SharpaRobotics's profile picture

Sharpa

5,539 subscribers

13,772 Aufrufe • vor 2 Monaten •via X (Twitter)

LIVE

Anya Rossi• Live Now

Private livecam show

0 Kommentare

Keine Kommentare verfügbar

Kommentare vom Original-Post werden hier angezeigt

Ähnliche Videos

15–30 minutes of real-world robot data. That's now enough to go from sim-to-real failure to working robot. Let’s see… You train a robot in simulation. You deploy it in the real world. It fails. The physics don't match. So you try to fine-tune it with real data, but… you never have enough real data, and the fine-tuning breaks everything the simulation taught it. SimDist fixes this with one key decision: don't transfer the policy. Transfer the world model. Keep the reward and value knowledge from simulation frozen. Only update the part that's actually wrong, how the robot predicts physics. Now the robot doesn't have to relearn the entire task in the real world. It already knows what success looks like. It just needs to correct its understanding of how the real world moves. The part that makes this work: they also trained on failures and recoveries; not just perfect demonstrations. Without that, the planner finds the gaps and exploits them. With it, the robot can tell a good future from a bad one. That's all it needs. Results on peg insertion, table leg assembly, locomotion on slippery and uneven surfaces. Tasks that require precision, force, and quick reaction. Thanks for sharing, Tyler Westenbroek ([Tyler Westenbroek]. Interactive visualization + paper: ——- Weekly robotics and AI insights. Subscribe free:
0:41

15–30 minutes of real-world robot data. That's now enough to go from sim-to-real failure to working robot. Let’s see… You train a robot in simulation. You deploy it in the real world. It fails. The physics don't match. So you try to fine-tune it with real data, but… you never have enough real data, and the fine-tuning breaks everything the simulation taught it. SimDist fixes this with one key decision: don't transfer the policy. Transfer the world model. Keep the reward and value knowledge from simulation frozen. Only update the part that's actually wrong, how the robot predicts physics. Now the robot doesn't have to relearn the entire task in the real world. It already knows what success looks like. It just needs to correct its understanding of how the real world moves. The part that makes this work: they also trained on failures and recoveries; not just perfect demonstrations. Without that, the planner finds the gaps and exploits them. With it, the robot can tell a good future from a bad one. That's all it needs. Results on peg insertion, table leg assembly, locomotion on slippery and uneven surfaces. Tasks that require precision, force, and quick reaction. Thanks for sharing, Tyler Westenbroek ([Tyler Westenbroek]. Interactive visualization + paper: ——- Weekly robotics and AI insights. Subscribe free:

Ilir Aliu

14,222 Aufrufe • vor 3 Monaten

Touch alone isn’t enough. 🖐️ For robotics, tactile intelligence truly levels up when touch gains spatial meaning. At #ICRA2026, we dove into the core concept behind SaTA: Spatially-anchored Tactile Awareness for robust, dexterous manipulation. Read it here: The challenge is fundamental: a robot shouldn’t just register a touch, it needs to understand exactly where that contact occurs relative to its fingers, joints, and overall hand structure. This is the missing link that turns raw data into precise, real-time adjustments during manipulation. Why does this matter? Because the most complex part of any manipulation task happens when vision is at its least reliable, that final millimeter before insertion, sliding, gripping, or fine-tuning. At Sharpa, this is exactly why we’re building tactile hands and tactile AI in tandem. 🚀 #Sharpa #Robotics #EmbodiedAI #TactileIntelligence #ICRA2026 #DexterousManipulation 📷
0:54

Touch alone isn’t enough. 🖐️ For robotics, tactile intelligence truly levels up when touch gains spatial meaning. At #ICRA2026, we dove into the core concept behind SaTA: Spatially-anchored Tactile Awareness for robust, dexterous manipulation. Read it here: The challenge is fundamental: a robot shouldn’t just register a touch, it needs to understand exactly where that contact occurs relative to its fingers, joints, and overall hand structure. This is the missing link that turns raw data into precise, real-time adjustments during manipulation. Why does this matter? Because the most complex part of any manipulation task happens when vision is at its least reliable, that final millimeter before insertion, sliding, gripping, or fine-tuning. At Sharpa, this is exactly why we’re building tactile hands and tactile AI in tandem. 🚀 #Sharpa #Robotics #EmbodiedAI #TactileIntelligence #ICRA2026 #DexterousManipulation 📷

Sharpa

15,168 Aufrufe • vor 3 Monaten

Another win for open-source robotics! 🔥 Hugging Face just released a fully open-source humanoid robot, and you can build one for $2,500. I'm a huge advocate of open-source in robotics space. Why? Robotics is too hard to solve alone. So lowering the entry point is crucial. Also, ROS proved open ecosystems create lasting industry standards. So let's have a look at the new member of humanoids family! LeRobot Humanoid is a 3D-printed, bipedal humanoid platform designed for real robot learning experiments. It's a complete and, buildable robot Anddddd yes, it's a full stack release: → Hardware files, bill of materials and full assembly documentation → Simulation assets and training environments → Runtime tools for calibration and real-world control → Sim-to-real identification pipeline When a part breaks, reprint it. When a design choice doesn't work, modify it and test again. The whole point is fast iteration on real hardware, not treating the robot as a fixed artefact. The timing is not a coincidence. As foundation models for robotics mature, the field desperately needs open physical platforms to train, validate and deploy them on. clem 🤗 LET'S GO!!! Build it here: ~~ ♻️ Join the weekly robotics newsletter, and never miss any news →
0:20

Another win for open-source robotics! 🔥 Hugging Face just released a fully open-source humanoid robot, and you can build one for $2,500. I'm a huge advocate of open-source in robotics space. Why? Robotics is too hard to solve alone. So lowering the entry point is crucial. Also, ROS proved open ecosystems create lasting industry standards. So let's have a look at the new member of humanoids family! LeRobot Humanoid is a 3D-printed, bipedal humanoid platform designed for real robot learning experiments. It's a complete and, buildable robot Anddddd yes, it's a full stack release: → Hardware files, bill of materials and full assembly documentation → Simulation assets and training environments → Runtime tools for calibration and real-world control → Sim-to-real identification pipeline When a part breaks, reprint it. When a design choice doesn't work, modify it and test again. The whole point is fast iteration on real hardware, not treating the robot as a fixed artefact. The timing is not a coincidence. As foundation models for robotics mature, the field desperately needs open physical platforms to train, validate and deploy them on. clem 🤗 LET'S GO!!! Build it here: ~~ ♻️ Join the weekly robotics newsletter, and never miss any news →

Lukas Ziegler

20,457 Aufrufe • vor 3 Monaten

Robots need to feel the world to operate in it. Most manipulation policies today are tactile-blind. They either cannot interpret high-frequency tactile signals or treat them as a static channel. And the field lacks enough touch-rich datasets to train tactile-reactive policies at scale. T-Rex was built to answer both. Dantong Niu and team, advised by Jim Fan, Fei-Fei Li, Jitendra MALIK, Pieter Abbeel trevordarrell have tested whether a robot policy can react to high-frequency tactile signals the way human hands do, without giving up the generalization power of modern VLAs. The result: 65% average success rate across 12 real-world tasks. +30 absolute points over the strongest baseline. In one year, tactile VLAs have gone from promising to outperforming non-tactile baselines like pi0.5 on dexterous tasks. #Robotics #SharpaWave #Sharpa #EmbodiedAI #DexterousManipulation #TactileSensing #RobotLearning Project link:
0:20

Robots need to feel the world to operate in it. Most manipulation policies today are tactile-blind. They either cannot interpret high-frequency tactile signals or treat them as a static channel. And the field lacks enough touch-rich datasets to train tactile-reactive policies at scale. T-Rex was built to answer both. Dantong Niu and team, advised by Jim Fan, Fei-Fei Li, Jitendra MALIK, Pieter Abbeel trevordarrell have tested whether a robot policy can react to high-frequency tactile signals the way human hands do, without giving up the generalization power of modern VLAs. The result: 65% average success rate across 12 real-world tasks. +30 absolute points over the strongest baseline. In one year, tactile VLAs have gone from promising to outperforming non-tactile baselines like pi0.5 on dexterous tasks. #Robotics #SharpaWave #Sharpa #EmbodiedAI #DexterousManipulation #TactileSensing #RobotLearning Project link:

Sharpa

12,931 Aufrufe • vor 2 Monaten

What if a dexterous robot could learn a 20+ step chemistry experiment without a single on-robot training demo? Meet TwinDEX: a pair of co-designed, three-finger, nine-DoF dexterous manipulation interface: one wearable for data collection, one for robot deployment. The twinned design shares identical kinematics, contact surfaces, visual appearance, and sensors across collection and deployment — keeping observations and actions aligned end to end. Trained from scratch on only a few hundred wearable demonstrations - with zero on-robot training or intervention data - TwinDEX completed a standardized chemistry experiment involving tool switches, fine force control, and bimanual coordination. Robot-free data showed comparable learning efficiency on the multi-task evaluation, TwinDEX delivered 5.3 times effective throughput than on-robot teleoperation. TwinDEX demonstrates that high-quality robot-free data can fully substitute for on-robot teleoperation data on challenging dexterous tasks — removing the dependency on real-robot hardware that has been the central bottleneck to scaling dexterous manipulation data. This was the proof-it phase. Now comes scale: what emerges at tens of thousands, or millions, of episodes? Watch the demo and read the technical blog: #TwinDEX #Robotics #EmbodiedAI #DexterousManipulation
4:31

What if a dexterous robot could learn a 20+ step chemistry experiment without a single on-robot training demo? Meet TwinDEX: a pair of co-designed, three-finger, nine-DoF dexterous manipulation interface: one wearable for data collection, one for robot deployment. The twinned design shares identical kinematics, contact surfaces, visual appearance, and sensors across collection and deployment — keeping observations and actions aligned end to end. Trained from scratch on only a few hundred wearable demonstrations - with zero on-robot training or intervention data - TwinDEX completed a standardized chemistry experiment involving tool switches, fine force control, and bimanual coordination. Robot-free data showed comparable learning efficiency on the multi-task evaluation, TwinDEX delivered 5.3 times effective throughput than on-robot teleoperation. TwinDEX demonstrates that high-quality robot-free data can fully substitute for on-robot teleoperation data on challenging dexterous tasks — removing the dependency on real-robot hardware that has been the central bottleneck to scaling dexterous manipulation data. This was the proof-it phase. Now comes scale: what emerges at tens of thousands, or millions, of episodes? Watch the demo and read the technical blog: #TwinDEX #Robotics #EmbodiedAI #DexterousManipulation

X Square Robot

1,340,894 Aufrufe • vor 7 Tagen

Humanoids finally move like humans… and can do more than copy. [Details + demos in thread 👇] A new framework, BeyondMimic, shows how to learn naturalistic whole-body control from human motion. But then goes further by composing those skills into versatile, zero-shot policies. Two key pieces make it work: •Motion tracking pipeline: robustly turns human mocap into dynamic robot motions (jump spins, sprinting, cartwheels) on real hardware •Guided diffusion policy: distills motion primitives and enables zero-shot downstream tasks with simple cost functions The result: ✅ State-of-the-art motion quality from mocap tracking ✅ Versatile task execution; waypoint navigation, joystick teleop, obstacle avoidance ✅ Stable sim-to-real transfer for diverse humanoid skills Thank you for sharing, Qiayuan Liao! 📍 Paper: Video:
0:42

Humanoids finally move like humans… and can do more than copy. [Details + demos in thread 👇] A new framework, BeyondMimic, shows how to learn naturalistic whole-body control from human motion. But then goes further by composing those skills into versatile, zero-shot policies. Two key pieces make it work: •Motion tracking pipeline: robustly turns human mocap into dynamic robot motions (jump spins, sprinting, cartwheels) on real hardware •Guided diffusion policy: distills motion primitives and enables zero-shot downstream tasks with simple cost functions The result: ✅ State-of-the-art motion quality from mocap tracking ✅ Versatile task execution; waypoint navigation, joystick teleop, obstacle avoidance ✅ Stable sim-to-real transfer for diverse humanoid skills Thank you for sharing, Qiayuan Liao! 📍 Paper: Video:

Ilir Aliu - eu/acc

72,522 Aufrufe • vor 1 Jahr

I’m thrilled to announce that we just released GraspGen, a multi-year project we have been cooking at NVIDIA Robotics 🚀 GraspGen: A Diffusion-Based Framework for 6-DOF Grasping Grasping is a foundational challenge in robotics 🤖 — whether for industrial picking or general-purpose humanoids. VLA + real data collection is all the rage now but is expensive and scales poorly for this task. For every new gripper and/or scene, you’ll have to recollect the dataset in this paradigm for the best perf. 💡Key Idea: Since grasping is such a well-defined task in simulation - why can’t we just scale synthetic data generation and train a generative model for grasping? By embracing modularity and standardized grasp formats, we can make this a turnkey technology that works zero-shot for multiple settings. GraspGen is a modular framework for diffusion-based 6-DOF grasp generation that scales across embodiment types, observability conditions, clutter, task complexity. Key Features: ✅ Multi-embodiment support: suction, parallel-jaw, and multi-fingered grippers ✅ Generalization to partial + complete 3D point clouds ✅ Generalization to single-objects + cluttered scenes ✅ Modular design uses other robotics modules and foundation models (SAM2, cuRobo, FoundationStereo, FoundationPose). This allows GraspGen to focus on only one thing - grasp generation ✅ Training recipe: grasp discriminator is trained with On-Generator data from the diffusion model - so that it learns to correct the mistakes (if any) of the diffusion generator ✅ Real-time performance (~20 Hz) before any GPU acceleration; low memory footprint 📊 Results: • SOTA on the FetchBench [Han et al. CoRL 2024] benchmark • Zero-shot sim-to-real transfer on unknown objects and cluttered scenes • Dataset of 53M simulated grasps across 8K objects from Objaverse 📄 arXiv: 🌐 Website: 💻 Code: A huge thank you to everyone involved in this journey — excited to see what the community builds on top of it! Joint work with Clemens Eppner , Balakumar Sundaralingam , Yu-Wei, Jun Yamada Wentao Yuan and other collaborators #robotics #diffusionmodels #physicalAI #simtoreal
0:39

I’m thrilled to announce that we just released GraspGen, a multi-year project we have been cooking at NVIDIA Robotics 🚀 GraspGen: A Diffusion-Based Framework for 6-DOF Grasping Grasping is a foundational challenge in robotics 🤖 — whether for industrial picking or general-purpose humanoids. VLA + real data collection is all the rage now but is expensive and scales poorly for this task. For every new gripper and/or scene, you’ll have to recollect the dataset in this paradigm for the best perf. 💡Key Idea: Since grasping is such a well-defined task in simulation - why can’t we just scale synthetic data generation and train a generative model for grasping? By embracing modularity and standardized grasp formats, we can make this a turnkey technology that works zero-shot for multiple settings. GraspGen is a modular framework for diffusion-based 6-DOF grasp generation that scales across embodiment types, observability conditions, clutter, task complexity. Key Features: ✅ Multi-embodiment support: suction, parallel-jaw, and multi-fingered grippers ✅ Generalization to partial + complete 3D point clouds ✅ Generalization to single-objects + cluttered scenes ✅ Modular design uses other robotics modules and foundation models (SAM2, cuRobo, FoundationStereo, FoundationPose). This allows GraspGen to focus on only one thing - grasp generation ✅ Training recipe: grasp discriminator is trained with On-Generator data from the diffusion model - so that it learns to correct the mistakes (if any) of the diffusion generator ✅ Real-time performance (~20 Hz) before any GPU acceleration; low memory footprint 📊 Results: • SOTA on the FetchBench [Han et al. CoRL 2024] benchmark • Zero-shot sim-to-real transfer on unknown objects and cluttered scenes • Dataset of 53M simulated grasps across 8K objects from Objaverse 📄 arXiv: 🌐 Website: 💻 Code: A huge thank you to everyone involved in this journey — excited to see what the community builds on top of it! Joint work with Clemens Eppner , Balakumar Sundaralingam , Yu-Wei, Jun Yamada Wentao Yuan and other collaborators #robotics #diffusionmodels #physicalAI #simtoreal

Adithya Murali

24,296 Aufrufe • vor 1 Jahr

Physics AI 🌊 Jensen just unveiled the NVIDIA Isaac GR00T Reference Humanoid Robot at GTC Taipei--the world’s first open-source humanoid robot reference design Built on Jetson Thor edge computing and the Isaac GR00T open platform, it features a Unitree H2 Plus body with Sharpa five-finger dexterous hands and 75 degrees of freedom. Even before Physical AI truly explodes, NVIDIA has built a complete end-to-end full-stack infrastructure: >chips (Jetson Thor + Blackwell), >synthetic data (GR00T-Dreams), >world models (Cosmos), >foundation models (GR00T N-series VLA), >high-fidelity simulation platforms (Isaac Sim + Isaac Lab + Omniverse). Dozens of humanoid robot companies ,including Agility Robotics, Aglie robotics,Boston Dynamics, Figure, 1X, NEURA Robotics, XPENG Robotics, and more are already running large-scale daily simulation training on this platform, rapidly iterating perception, reasoning, and actions in virtual environments, speeding up the jump from lab prototypes to real factory deployment. By providing an open reference design and full-stack tools, NVIDIA has dramatically lowered the barrier, letting every robotics team join the Physical AI wave with low cost and high efficiency.
4:31

Physics AI 🌊 Jensen just unveiled the NVIDIA Isaac GR00T Reference Humanoid Robot at GTC Taipei--the world’s first open-source humanoid robot reference design Built on Jetson Thor edge computing and the Isaac GR00T open platform, it features a Unitree H2 Plus body with Sharpa five-finger dexterous hands and 75 degrees of freedom. Even before Physical AI truly explodes, NVIDIA has built a complete end-to-end full-stack infrastructure: >chips (Jetson Thor + Blackwell), >synthetic data (GR00T-Dreams), >world models (Cosmos), >foundation models (GR00T N-series VLA), >high-fidelity simulation platforms (Isaac Sim + Isaac Lab + Omniverse). Dozens of humanoid robot companies ,including Agility Robotics, Aglie robotics,Boston Dynamics, Figure, 1X, NEURA Robotics, XPENG Robotics, and more are already running large-scale daily simulation training on this platform, rapidly iterating perception, reasoning, and actions in virtual environments, speeding up the jump from lab prototypes to real factory deployment. By providing an open reference design and full-stack tools, NVIDIA has dramatically lowered the barrier, letting every robotics team join the Physical AI wave with low cost and high efficiency.

CyberRobo

17,092 Aufrufe • vor 3 Monaten

🚨 BREAKING: ABB Robotics + NVIDIA close the sim-to-real gap with 99% accuracy! 👾 ABB Robotics is integrating NVIDIA Omniverse libraries into RobotStudio to deliver physical AI for industry, closing the gap from virtual training to real-world deployment with up to 99% accuracy. RobotStudio HyperReality, available second half of 2026, will fundamentally change how quickly manufacturers can scale production: reducing costs by up to 40%, accelerating time-to-market by 50%, and cutting setup and commissioning times by up to 80%. For decades, the deficit between simulation accuracy and real-world lighting, materials, and environments has limited manufacturers' ability to design advanced manufacturing processes in the virtual world. The only robot manufacturer with a virtual controller running the same firmware as the hardware, ensuring near-perfect correlation between simulation and real-world performance. The system uses physically accurate simulations and foundation models endlessly optimized with real-world data feedback. These models can train any number of ABB robots anywhere in the world with industrial-grade reliability. Foxconn is using RobotStudio HyperReality for consumer electronics assembly. Assembly robots are trained virtually using synthetic data to perfect multiple production processes across various scenarios, then moved to production lines with 99% accuracy. This eliminates physical training and tests, reducing setup times and costs. Workr is demonstrating AI-powered robotic systems at NVIDIA GTC 2026. Built on ABB technology, trained with synthetic data using NVIDIA Omniverse, deployed without operators needing programming knowledge . 🚨 I’ll be onsite in San Jose during GTC 2026, and will be showing all the cool stuff that ABB Robotics prepared this year! Can’t wait! 🫡 ~~ ♻️ Join the weekly robotics newsletter, and never miss any news →
0:48

🚨 BREAKING: ABB Robotics + NVIDIA close the sim-to-real gap with 99% accuracy! 👾 ABB Robotics is integrating NVIDIA Omniverse libraries into RobotStudio to deliver physical AI for industry, closing the gap from virtual training to real-world deployment with up to 99% accuracy. RobotStudio HyperReality, available second half of 2026, will fundamentally change how quickly manufacturers can scale production: reducing costs by up to 40%, accelerating time-to-market by 50%, and cutting setup and commissioning times by up to 80%. For decades, the deficit between simulation accuracy and real-world lighting, materials, and environments has limited manufacturers' ability to design advanced manufacturing processes in the virtual world. The only robot manufacturer with a virtual controller running the same firmware as the hardware, ensuring near-perfect correlation between simulation and real-world performance. The system uses physically accurate simulations and foundation models endlessly optimized with real-world data feedback. These models can train any number of ABB robots anywhere in the world with industrial-grade reliability. Foxconn is using RobotStudio HyperReality for consumer electronics assembly. Assembly robots are trained virtually using synthetic data to perfect multiple production processes across various scenarios, then moved to production lines with 99% accuracy. This eliminates physical training and tests, reducing setup times and costs. Workr is demonstrating AI-powered robotic systems at NVIDIA GTC 2026. Built on ABB technology, trained with synthetic data using NVIDIA Omniverse, deployed without operators needing programming knowledge . 🚨 I’ll be onsite in San Jose during GTC 2026, and will be showing all the cool stuff that ABB Robotics prepared this year! Can’t wait! 🫡 ~~ ♻️ Join the weekly robotics newsletter, and never miss any news →

Lukas Ziegler

22,482 Aufrufe • vor 6 Monaten

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

909,131 Aufrufe • vor 2 Jahren

Today, we give robots a /skills library that self-evolves and compounds indefinitely! Introducing ASPIRE: a robot solving its 100th task is no longer as clueless as solving its first. Coding agents observe multimodal sensory traces from simulation and real robots, launch an evolutionary search over control programs, and distill the best know-how into an ever-expanding library. ASPIRE is a new type of continual learning: "training" is skill refinement instead of gradient descent. "Trained model" is a repo of sensorimotor skills instead of floating weights. “Distributed training” is a panel of agents each practicing a different skill instead of sharded minibatches. Here's the beauty: ASPIRE gives the tired terms "sim2real transfer" and "cross-embodiment transfer" a whole new meaning. Bridging the sim-to-real gap is notoriously brutal. An end-to-end policy has to swallow both the visual shift (sim looks toyish next to a real camera) and the subtle contact physics it never quite gets right. ASPIRE sidesteps the mess, because it doesn't ship pixels or weights across the gap, but ships the know-how. The robot still has to practice in the real world, not zero-shot, but it gets there way faster because it isn't rediscovering the strategy from scratch. Same for going single-arm to bimanual hardware, which usually requires new data and retraining from zero. ASPIRE achieves up to ~10x cut in "transfer learning” tokens (yes, tokens are the new unit of *training* compute ;) Check out our gallery of 150+ tasks and 90+ skills the robots taught themselves, all on the website! Kind of wild that we can ship the "learned weights" as an HTML page rather than a GGUF. We'll open-source the full stack so your own robot library starts compounding from ours! Deep dive in thread:
0:40

Today, we give robots a /skills library that self-evolves and compounds indefinitely! Introducing ASPIRE: a robot solving its 100th task is no longer as clueless as solving its first. Coding agents observe multimodal sensory traces from simulation and real robots, launch an evolutionary search over control programs, and distill the best know-how into an ever-expanding library. ASPIRE is a new type of continual learning: "training" is skill refinement instead of gradient descent. "Trained model" is a repo of sensorimotor skills instead of floating weights. “Distributed training” is a panel of agents each practicing a different skill instead of sharded minibatches. Here's the beauty: ASPIRE gives the tired terms "sim2real transfer" and "cross-embodiment transfer" a whole new meaning. Bridging the sim-to-real gap is notoriously brutal. An end-to-end policy has to swallow both the visual shift (sim looks toyish next to a real camera) and the subtle contact physics it never quite gets right. ASPIRE sidesteps the mess, because it doesn't ship pixels or weights across the gap, but ships the know-how. The robot still has to practice in the real world, not zero-shot, but it gets there way faster because it isn't rediscovering the strategy from scratch. Same for going single-arm to bimanual hardware, which usually requires new data and retraining from zero. ASPIRE achieves up to ~10x cut in "transfer learning” tokens (yes, tokens are the new unit of *training* compute ;) Check out our gallery of 150+ tasks and 90+ skills the robots taught themselves, all on the website! Kind of wild that we can ship the "learned weights" as an HTML page rather than a GGUF. We'll open-source the full stack so your own robot library starts compounding from ours! Deep dive in thread:

Jim Fan

211,649 Aufrufe • vor 2 Monaten

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

300,212 Aufrufe • vor 6 Monaten

NEW ROBOT BENCHMARK: If your robot can do Origami, it can do anything! Called The Robotic Origami Challenge, it is a dexterous-manipulation competition and benchmark held at IROS 2026, organized by 13 co-organizers with the Nippon Origami Association as judge and task curator -> pretty cool to have them on board imho. The evaluation consists of single task: a traditional Japanese paper airplane, in exactly six folds, from a 15×15 cm sheet of ≥60 gsm paper, on a competition-supplied standardized rig (bimanual arms + Sharpa Hands), both remotely and on-site. Teams bring policies, not hardware. An "Origami Grand Master" declares pass/fail on crease accuracy, structural fidelity, symmetry and paper integrity. Among passes, faster folds rank higher, with a 10-minute-per-attempt ceiling and flight explicitly irrelevant to the score. -> I find it interesting how they chose to evaluate the task. Quality is a binary pass or fail, therefore speed becomes the only thing graded here. Speed is currently the bottleneck in dexterous manipulation though, so this choice makes sense. I wonder whether there could be finer ways to grade the qaulity of such a creative task though. When it comes to data, registered teams get 500+ teleoperation episodes (six camera streams, 65-D joint state/action, 10-fingertip 6-axis tactile), an NVIDIA Isaac Sim environment with thin-shell paper physics (plastic creasing + fold memory), digital twins of every partner hand, and a remote eval lab (upload a policy, queue an eval, get scored). Still, I think it is a great dexterity benchmark the field badly needs, it supplies the hardware, an outside human judges, and the pass criterion is externally defined -> all three degrees of freedom are checked! Neutral measurement layer, here we go! The task is engineered to be un-gameable and to isolate pure dexterity. A known figure, exactly six folds, judged on creases, with flight explicitly declared irrelevant (the latter makes sense to me). Therefore, this underlines the goal to focus on dexterity, not task-selection or other strategies. I really like origami as an ideal controlled dexterity task: deterministic goal, deformable medium, sequential, bimanual, precision-bound. I am just not quite satisfied again by the binary pass or fail, I think quality of execution could be finely graded! But again, I understand this is not the goal yet. Also interesting to see the Sharpa Hands as de facto standard for everyone. Total land-grab that anoints Sharpa as the reference dexterous hand, also featured in Google's Gemini Robotics 2. By providing the hardware, the benchmark measures software while quietly making "good on Sharpa" the definition of good, and Sharpa gets real world data and feedback as a bonus. That's smart, the data flywheel starts spinning. The provided dataset is the richest tactile-manipulation corpus I have seen yet: 10-fingertip 6-axis tactile, plus plastic creasing and fold memory. 500+ teleop episodes with six camera streams, 65-D joint state/action, and ten fingertip 6-axis tactile sensors. The force/tactile channel are parts of the the benchmark's data, this is the first time I see this. Credits where it's due: organizers include Yang Gao, Noriaki Hirose, Steve Xie, Chris Paxton, Jiafei Duan, Michael Cho - Rbt/Acc, Michael Yuan, Haoquan Fang, and others.
0:36

NEW ROBOT BENCHMARK: If your robot can do Origami, it can do anything! Called The Robotic Origami Challenge, it is a dexterous-manipulation competition and benchmark held at IROS 2026, organized by 13 co-organizers with the Nippon Origami Association as judge and task curator -> pretty cool to have them on board imho. The evaluation consists of single task: a traditional Japanese paper airplane, in exactly six folds, from a 15×15 cm sheet of ≥60 gsm paper, on a competition-supplied standardized rig (bimanual arms + Sharpa Hands), both remotely and on-site. Teams bring policies, not hardware. An "Origami Grand Master" declares pass/fail on crease accuracy, structural fidelity, symmetry and paper integrity. Among passes, faster folds rank higher, with a 10-minute-per-attempt ceiling and flight explicitly irrelevant to the score. -> I find it interesting how they chose to evaluate the task. Quality is a binary pass or fail, therefore speed becomes the only thing graded here. Speed is currently the bottleneck in dexterous manipulation though, so this choice makes sense. I wonder whether there could be finer ways to grade the qaulity of such a creative task though. When it comes to data, registered teams get 500+ teleoperation episodes (six camera streams, 65-D joint state/action, 10-fingertip 6-axis tactile), an NVIDIA Isaac Sim environment with thin-shell paper physics (plastic creasing + fold memory), digital twins of every partner hand, and a remote eval lab (upload a policy, queue an eval, get scored). Still, I think it is a great dexterity benchmark the field badly needs, it supplies the hardware, an outside human judges, and the pass criterion is externally defined -> all three degrees of freedom are checked! Neutral measurement layer, here we go! The task is engineered to be un-gameable and to isolate pure dexterity. A known figure, exactly six folds, judged on creases, with flight explicitly declared irrelevant (the latter makes sense to me). Therefore, this underlines the goal to focus on dexterity, not task-selection or other strategies. I really like origami as an ideal controlled dexterity task: deterministic goal, deformable medium, sequential, bimanual, precision-bound. I am just not quite satisfied again by the binary pass or fail, I think quality of execution could be finely graded! But again, I understand this is not the goal yet. Also interesting to see the Sharpa Hands as de facto standard for everyone. Total land-grab that anoints Sharpa as the reference dexterous hand, also featured in Google's Gemini Robotics 2. By providing the hardware, the benchmark measures software while quietly making "good on Sharpa" the definition of good, and Sharpa gets real world data and feedback as a bonus. That's smart, the data flywheel starts spinning. The provided dataset is the richest tactile-manipulation corpus I have seen yet: 10-fingertip 6-axis tactile, plus plastic creasing and fold memory. 500+ teleop episodes with six camera streams, 65-D joint state/action, and ten fingertip 6-axis tactile sensors. The force/tactile channel are parts of the the benchmark's data, this is the first time I see this. Credits where it's due: organizers include Yang Gao, Noriaki Hirose, Steve Xie, Chris Paxton, Jiafei Duan, Michael Cho - Rbt/Acc, Michael Yuan, Haoquan Fang, and others.

Léo

28,693 Aufrufe • vor 1 Monat

A policy that teaches robot hands to touch things the way humans do... not just grab and move, but feel and adjust in real time. Robot manipulation research often stops at picking up objects and placing them. CGP goes further: it handles tasks like opening jars, flipping objects in-hand, wiping dishes, and grasping fragile eggs, the kind of dexterous, contact-rich skills that require constant micro-adjustments based on what the fingers are actually feeling. The robot doesn't just see what it's doing; it predicts what contact should feel like at each step, then checks whether reality matches the prediction. If a finger is slipping, the policy knows before the object drops. Works on real robot hands (both 4-finger and 5-finger designs) with tactile sensors embedded in the fingertips Robust to visual distractions! The robot keeps flipping a box correctly even when the camera view is disrupted, because it's grounding decisions in touch, not just vision. Baseline policies without contact grounding fail in predictable ways: slipping mid-task, incomplete motions, loss of grasp, CGP avoids these This is a meaningful step toward robots that can handle the physical world with the kind of reliable, adaptive grip that humans take for granted. Relevant for manufacturing, logistics, assistive robotics, and anywhere fragile or irregular objects need to be handled carefully. Published at RSS 2026, developed with Meta Reality Labs Research. Thanks for sharing, Zhengtong Xu / Zhengtong Xu ——- Weekly robotics and AI insights. Subscribe free:
0:36

A policy that teaches robot hands to touch things the way humans do... not just grab and move, but feel and adjust in real time. Robot manipulation research often stops at picking up objects and placing them. CGP goes further: it handles tasks like opening jars, flipping objects in-hand, wiping dishes, and grasping fragile eggs, the kind of dexterous, contact-rich skills that require constant micro-adjustments based on what the fingers are actually feeling. The robot doesn't just see what it's doing; it predicts what contact should feel like at each step, then checks whether reality matches the prediction. If a finger is slipping, the policy knows before the object drops. Works on real robot hands (both 4-finger and 5-finger designs) with tactile sensors embedded in the fingertips Robust to visual distractions! The robot keeps flipping a box correctly even when the camera view is disrupted, because it's grounding decisions in touch, not just vision. Baseline policies without contact grounding fail in predictable ways: slipping mid-task, incomplete motions, loss of grasp, CGP avoids these This is a meaningful step toward robots that can handle the physical world with the kind of reliable, adaptive grip that humans take for granted. Relevant for manufacturing, logistics, assistive robotics, and anywhere fragile or irregular objects need to be handled carefully. Published at RSS 2026, developed with Meta Reality Labs Research. Thanks for sharing, Zhengtong Xu / Zhengtong Xu ——- Weekly robotics and AI insights. Subscribe free:

Ilir Aliu

12,769 Aufrufe • vor 3 Monaten

Tim Ferriss on the dangerous trap hiding inside self-help: Most people approach self-improvement the same way someone might prepare to play soccer, except they never actually get on the field. Tim describes this pattern in striking detail: "You want to play soccer but first you're going to read all the textbooks and get a master's degree and PhD in soccer and then you're going to practice dribbling and penalty shots and so on by yourself and you want to become as perfect a player as possible by yourself before you ever actually get on the field and play the game of soccer." The result? You begin to believe that practising alone is the same as playing the game. This is the hidden danger Tim calls the self-help trap, the implicit belief that you must fix yourself, do the work, and polish yourself to readiness before you can meaningfully engage with other people, relationships, or family. The problem is that it never ends. There's always another edge to smooth, another flaw to address. The self becomes a project with no completion date. As Tim puts it: "You're always polishing this self and it can become this real recursive dangerous trap, this fixation on the self." The real game, relationships, family, community is learned by playing, not by preparing to play. The friction, the discomfort, the messiness of showing up imperfectly with other people is the development. You can't practise your way into readiness for it in isolation. The irony of self-help is that taken too far, it keeps you away from the very thing you're supposedly preparing for.
1:14

Tim Ferriss on the dangerous trap hiding inside self-help: Most people approach self-improvement the same way someone might prepare to play soccer, except they never actually get on the field. Tim describes this pattern in striking detail: "You want to play soccer but first you're going to read all the textbooks and get a master's degree and PhD in soccer and then you're going to practice dribbling and penalty shots and so on by yourself and you want to become as perfect a player as possible by yourself before you ever actually get on the field and play the game of soccer." The result? You begin to believe that practising alone is the same as playing the game. This is the hidden danger Tim calls the self-help trap, the implicit belief that you must fix yourself, do the work, and polish yourself to readiness before you can meaningfully engage with other people, relationships, or family. The problem is that it never ends. There's always another edge to smooth, another flaw to address. The self becomes a project with no completion date. As Tim puts it: "You're always polishing this self and it can become this real recursive dangerous trap, this fixation on the self." The real game, relationships, family, community is learned by playing, not by preparing to play. The friction, the discomfort, the messiness of showing up imperfectly with other people is the development. You can't practise your way into readiness for it in isolation. The irony of self-help is that taken too far, it keeps you away from the very thing you're supposedly preparing for.

Kevin Tanaka

16,961 Aufrufe • vor 5 Monaten

This is one-shot assembly: you show examples of what to build, and the robot just does it. (see original post: To share more on how this works, the robot is controlled in real time by a neural network that takes in video pixels and outputs 100Hz actions. The video below is part of the raw input passed directly into the model. I also like this view (at 1x speed) because it shows more of the (I think very cool) subtle moments of dexterity near the fingertips 👌 One-shot assembly seemed like a dream even just a year ago — it's not easy. It requires both the high-level reasoning of "what to build" (recognizing the geometry of the structures presented by the human), and the low-level visuomotor control of "how to build it" (purposefully re-orienting individual pieces and nudging them together in place). While possible to manually engineer a complex system for this (e.g. w/ hierarchical control, or explicit state representations), we were curious if our own Foundation model could do it all end-to-end with just some post-training data. Surprisingly, it just worked. Nothing about the recipe is substantially different than any other demo we’ve run in the past, and we’re excited about its implications on model capabilities: • On contextual reasoning, these models can (i) attend to task-related pixels in the peripheral view of the video inputs, and (ii) retain this knowledge in-context while ignoring irrelevant background. This is useful for generalizing to a wide range of real workflows: e.g. paying attention to what’s coming down the conveyor line, or glancing at the instructions displayed on a nearby monitor. • On dexterity, these models can produce contact-rich "commonsense" behaviors that can be difficult to pre-program or write language instructions for e.g. rolling a brick slightly to align its studs against the bottom of another, re-grasping to get a better grip or to move out of the way before a forceful press, or gently pushing the corners of a brick against the mat to rotate it in hand and stand it up vertically (i.e. extrinsic dexterity). These aspects work together to form a capability that resembles fast adaptation — a hallmark of intelligence, relevant for real use cases. This has also expanded my own perspective on what's possible with robot learning, using a recipe that's repeatable for many more skills. This milestone stands on top of the solid technical foundations we’ve built here at Generalist: hardcore controls & hardware, all in-house built models, and a data engine that "just works." We're a small group of hyper-focused engineers, and hands-down the highest talent-density team I’ve ever worked with. We're accelerating and scaling aggressively towards unlocking next-generation robot intelligence. Building Legos is just one example, and it's clear to me that we're headed towards a future where robots can do just about anything we want them to. Its coming, and we're going to make it happen.
1:01

This is one-shot assembly: you show examples of what to build, and the robot just does it. (see original post: To share more on how this works, the robot is controlled in real time by a neural network that takes in video pixels and outputs 100Hz actions. The video below is part of the raw input passed directly into the model. I also like this view (at 1x speed) because it shows more of the (I think very cool) subtle moments of dexterity near the fingertips 👌 One-shot assembly seemed like a dream even just a year ago — it's not easy. It requires both the high-level reasoning of "what to build" (recognizing the geometry of the structures presented by the human), and the low-level visuomotor control of "how to build it" (purposefully re-orienting individual pieces and nudging them together in place). While possible to manually engineer a complex system for this (e.g. w/ hierarchical control, or explicit state representations), we were curious if our own Foundation model could do it all end-to-end with just some post-training data. Surprisingly, it just worked. Nothing about the recipe is substantially different than any other demo we’ve run in the past, and we’re excited about its implications on model capabilities: • On contextual reasoning, these models can (i) attend to task-related pixels in the peripheral view of the video inputs, and (ii) retain this knowledge in-context while ignoring irrelevant background. This is useful for generalizing to a wide range of real workflows: e.g. paying attention to what’s coming down the conveyor line, or glancing at the instructions displayed on a nearby monitor. • On dexterity, these models can produce contact-rich "commonsense" behaviors that can be difficult to pre-program or write language instructions for e.g. rolling a brick slightly to align its studs against the bottom of another, re-grasping to get a better grip or to move out of the way before a forceful press, or gently pushing the corners of a brick against the mat to rotate it in hand and stand it up vertically (i.e. extrinsic dexterity). These aspects work together to form a capability that resembles fast adaptation — a hallmark of intelligence, relevant for real use cases. This has also expanded my own perspective on what's possible with robot learning, using a recipe that's repeatable for many more skills. This milestone stands on top of the solid technical foundations we’ve built here at Generalist: hardcore controls & hardware, all in-house built models, and a data engine that "just works." We're a small group of hyper-focused engineers, and hands-down the highest talent-density team I’ve ever worked with. We're accelerating and scaling aggressively towards unlocking next-generation robot intelligence. Building Legos is just one example, and it's clear to me that we're headed towards a future where robots can do just about anything we want them to. Its coming, and we're going to make it happen.

Andy Zeng

49,443 Aufrufe • vor 11 Monaten

CHINA JUST SOLVED THE PROBLEM THAT'S BEEN BREAKING ROBOT AI FOR A DECADE. and the fix wasn't a smarter model. for years, every robot AI failure got the same diagnosis. the model isn't smart enough. so everyone scaled intelligence. bigger models. more parameters. better reasoning. AGIBOT asked a different question: what if the reasoning was never the problem? there's a gap that runs through every traditional robot AI system. reasoning on one side & motor commands on the other. the brain decides but the body executes something different, because thinking and moving were never actually connected. GO-2 fixes this by reasoning INSIDE the action space, not above it. before moving, it runs a complete mental simulation of every step - like a basketball player mentally tracing the arc of a shot before releasing the ball. watch the demo and you'll see exactly what this means. the robot works through a task queue autonomously. classify toiletries. upright the drink bottle. place headphones in the leather box. mid-execution, a new instruction drops: "my phone's missing. help me find it." it doesn't pause. doesn't reset. it processes the new task and keeps moving. that's not a scripted sequence. that's real-time instruction following on top of an active task queue. that one architectural change is where the numbers come from. > #1 on LIBERO across Spatial, Object, Goal, and Long tasks → 98.5% average success > 86.6% zero-shot accuracy in active disturbance environments > 47.4 on VLABench → best-in-class on objects and textures it's never seen before > 82.9% success trained on simulation only, tested on real hardware sim-to-real is the graveyard of robotics research. models trained in simulation collapse the moment they touch the real world. 82.9% means that graveyard just got a lot smaller. it holds because of how GO-2 trains. deliberately fed imperfect reasoning conditions, then trained to execute robustly anyway. not a researcher assumption. a design decision from a team that ships hardware and knows exactly what breaks. then there's the infrastructure layer. Genie Studio. fleet-wide data collection. cloud training. online post-training in live environments. 10x improvement in training efficiency. task startup reduced to minutes. 2-4x better success rates with 50%+ less data. the model gets smarter every time a robot fails in the field. this isn't a benchmark story. it's a compounding moat. dual CVPR 2026 + ACL 2026 acceptance. computer vision AND natural language processing. top conferences. simultaneously. that doesn't happen with incremental research. the US-China robotics race has been framed as a compute race. a model quality race. it was always an execution race. the robot that wins won't be the smartest one in the lab. it'll be the most reliable one on the floor. full breakdown: is execution reliability the real bottleneck, or are we still underestimating how far reasoning needs to go?
2:55

CHINA JUST SOLVED THE PROBLEM THAT'S BEEN BREAKING ROBOT AI FOR A DECADE. and the fix wasn't a smarter model. for years, every robot AI failure got the same diagnosis. the model isn't smart enough. so everyone scaled intelligence. bigger models. more parameters. better reasoning. AGIBOT asked a different question: what if the reasoning was never the problem? there's a gap that runs through every traditional robot AI system. reasoning on one side & motor commands on the other. the brain decides but the body executes something different, because thinking and moving were never actually connected. GO-2 fixes this by reasoning INSIDE the action space, not above it. before moving, it runs a complete mental simulation of every step - like a basketball player mentally tracing the arc of a shot before releasing the ball. watch the demo and you'll see exactly what this means. the robot works through a task queue autonomously. classify toiletries. upright the drink bottle. place headphones in the leather box. mid-execution, a new instruction drops: "my phone's missing. help me find it." it doesn't pause. doesn't reset. it processes the new task and keeps moving. that's not a scripted sequence. that's real-time instruction following on top of an active task queue. that one architectural change is where the numbers come from. > #1 on LIBERO across Spatial, Object, Goal, and Long tasks → 98.5% average success > 86.6% zero-shot accuracy in active disturbance environments > 47.4 on VLABench → best-in-class on objects and textures it's never seen before > 82.9% success trained on simulation only, tested on real hardware sim-to-real is the graveyard of robotics research. models trained in simulation collapse the moment they touch the real world. 82.9% means that graveyard just got a lot smaller. it holds because of how GO-2 trains. deliberately fed imperfect reasoning conditions, then trained to execute robustly anyway. not a researcher assumption. a design decision from a team that ships hardware and knows exactly what breaks. then there's the infrastructure layer. Genie Studio. fleet-wide data collection. cloud training. online post-training in live environments. 10x improvement in training efficiency. task startup reduced to minutes. 2-4x better success rates with 50%+ less data. the model gets smarter every time a robot fails in the field. this isn't a benchmark story. it's a compounding moat. dual CVPR 2026 + ACL 2026 acceptance. computer vision AND natural language processing. top conferences. simultaneously. that doesn't happen with incremental research. the US-China robotics race has been framed as a compute race. a model quality race. it was always an execution race. the robot that wins won't be the smartest one in the lab. it'll be the most reliable one on the floor. full breakdown: is execution reliability the real bottleneck, or are we still underestimating how far reasoning needs to go?

Shruti

18,622 Aufrufe • vor 4 Monaten

Announcing DreamDojo: our open-source, interactive world model that takes robot motor controls and generates the future in pixels. No engine, no meshes, no hand-authored dynamics. It's Simulation 2.0. Time for robotics to take the bitter lesson pill. Real-world robot learning is bottlenecked by time, wear, safety, and resets. If we want Physical AI to move at pretraining speed, we need a simulator that adapts to pretraining scale with as little human engineering as possible. Our key insights: (1) human egocentric videos are a scalable source of first-person physics; (2) latent actions make them "robot-readable" across different hardware; (3) real-time inference unlocks live teleop, policy eval, and test-time planning *inside* a dream. We pre-train on 44K hours of human videos: cheap, abundant, and collected with zero robot-in-the-loop. Humans have already explored the combinatorics: we grasp, pour, fold, assemble, fail, retry—across cluttered scenes, shifting viewpoints, changing light, and hour-long task chains—at a scale no robot fleet could match. The missing piece: these videos have no action labels. So we introduce latent actions: a unified representation inferred directly from videos that captures "what changed between world states" without knowing the underlying hardware. This lets us train on any first-person video as if it came with motor commands attached. As a result, DreamDojo generalizes zero-shot to objects and environments never seen in any robot training set, because humans saw them first. Next, we post-train onto each robot to fit its specific hardware. Think of it as separating "how the world looks and behaves" from "how this particular robot actuates." The base model follows the general physical rules, then "snaps onto" the robot's unique mechanics. It's kind of like loading a new character and scene assets into Unreal Engine, but done through gradient descent and generalizes far beyond the post-training dataset. A world simulator is only useful if it runs fast enough to close the loop. We train a real-time version of DreamDojo that runs at 10 FPS, stable for over a minute of continuous rollout. This unlocks exciting possibilities: - Live teleoperation *inside* a dream. Connect a VR controller, stream actions into DreamDojo, and teleop a virtual robot in real time. We demo this on Unitree G1 with a PICO headset and one RTX 5090. - Policy evaluation. You can benchmark a policy checkpoint in DreamDojo instead of the real world. The simulated success rates strongly correlate with real-world results - accurate enough to rank checkpoints without burning a single motor. - Model-based planning. Sample multiple action proposals → simulate them all in parallel → pick the best future. Gains +17% real-world success out of the box on a fruit packing task. We open-source everything!! Weights, code, post-training dataset, eval set, and whitepaper with tons of details to reproduce. DreamDojo is based on NVIDIA Cosmos, which is open-weight too. 2026 is the year of World Models for physical AI. We want you to build with us. Happy scaling! Links in thread:
1:17

Announcing DreamDojo: our open-source, interactive world model that takes robot motor controls and generates the future in pixels. No engine, no meshes, no hand-authored dynamics. It's Simulation 2.0. Time for robotics to take the bitter lesson pill. Real-world robot learning is bottlenecked by time, wear, safety, and resets. If we want Physical AI to move at pretraining speed, we need a simulator that adapts to pretraining scale with as little human engineering as possible. Our key insights: (1) human egocentric videos are a scalable source of first-person physics; (2) latent actions make them "robot-readable" across different hardware; (3) real-time inference unlocks live teleop, policy eval, and test-time planning *inside* a dream. We pre-train on 44K hours of human videos: cheap, abundant, and collected with zero robot-in-the-loop. Humans have already explored the combinatorics: we grasp, pour, fold, assemble, fail, retry—across cluttered scenes, shifting viewpoints, changing light, and hour-long task chains—at a scale no robot fleet could match. The missing piece: these videos have no action labels. So we introduce latent actions: a unified representation inferred directly from videos that captures "what changed between world states" without knowing the underlying hardware. This lets us train on any first-person video as if it came with motor commands attached. As a result, DreamDojo generalizes zero-shot to objects and environments never seen in any robot training set, because humans saw them first. Next, we post-train onto each robot to fit its specific hardware. Think of it as separating "how the world looks and behaves" from "how this particular robot actuates." The base model follows the general physical rules, then "snaps onto" the robot's unique mechanics. It's kind of like loading a new character and scene assets into Unreal Engine, but done through gradient descent and generalizes far beyond the post-training dataset. A world simulator is only useful if it runs fast enough to close the loop. We train a real-time version of DreamDojo that runs at 10 FPS, stable for over a minute of continuous rollout. This unlocks exciting possibilities: - Live teleoperation *inside* a dream. Connect a VR controller, stream actions into DreamDojo, and teleop a virtual robot in real time. We demo this on Unitree G1 with a PICO headset and one RTX 5090. - Policy evaluation. You can benchmark a policy checkpoint in DreamDojo instead of the real world. The simulated success rates strongly correlate with real-world results - accurate enough to rank checkpoints without burning a single motor. - Model-based planning. Sample multiple action proposals → simulate them all in parallel → pick the best future. Gains +17% real-world success out of the box on a fruit packing task. We open-source everything!! Weights, code, post-training dataset, eval set, and whitepaper with tons of details to reproduce. DreamDojo is based on NVIDIA Cosmos, which is open-weight too. 2026 is the year of World Models for physical AI. We want you to build with us. Happy scaling! Links in thread:

Jim Fan

228,146 Aufrufe • vor 6 Monaten

I don't know if you really understand what these people just did... They built an AI that looks at a few casual real-world videos or photos and instantly turns them into a full 3D playground. You can MOVE THE CAMERA HOWEVER YOU WANT, change the lighting, move objects around, and even let physics play out like in the real world. View events from impossible angles because Atlas knows what was there. This is massive! Let's take robotics. Until now training robots meant collecting tons of expensive real-world data. Now a handful of recordings can become thousands of different simulated situations. Robots can practice the same task in new rooms, with different objects and lighting without anyone having to film it all again. For the industry it means we can train robots way faster and cheaper. Home robots, warehouse robots, whatever... they can learn new environments at scale instead of staying stuck in the lab. This is one of the first real bridges between “cool AI video models” and actual physical machines that have to work in the real world. Blog: The team behind World Labs: Fei-Fei Li, Justin Johnson, Ben Mildenhall and Yunzhu Li. The video is from David Pantera. Follow for more insights into robotics and AI. —— Weekly robotics and AI insights. Subscribe free:
0:49

I don't know if you really understand what these people just did... They built an AI that looks at a few casual real-world videos or photos and instantly turns them into a full 3D playground. You can MOVE THE CAMERA HOWEVER YOU WANT, change the lighting, move objects around, and even let physics play out like in the real world. View events from impossible angles because Atlas knows what was there. This is massive! Let's take robotics. Until now training robots meant collecting tons of expensive real-world data. Now a handful of recordings can become thousands of different simulated situations. Robots can practice the same task in new rooms, with different objects and lighting without anyone having to film it all again. For the industry it means we can train robots way faster and cheaper. Home robots, warehouse robots, whatever... they can learn new environments at scale instead of staying stuck in the lab. This is one of the first real bridges between “cool AI video models” and actual physical machines that have to work in the real world. Blog: The team behind World Labs: Fei-Fei Li, Justin Johnson, Ben Mildenhall and Yunzhu Li. The video is from David Pantera. Follow for more insights into robotics and AI. —— Weekly robotics and AI insights. Subscribe free:

Ilir Aliu

270,398 Aufrufe • vor 7 Tagen

Sunday Robotics' Tony Zhao runs a fully autonomous robot demo live and unscripted, and it corrects itself every time you mess with it: brexton: "I want to confirm for the audience right now that this is fully autonomous. There's no teleoperation, no one behind the curtain, and this is real time. And we did not prepare this demo before. Why do you think seeing real-time demos is so rare for robotics? What the audience is seeing right now is not that common." Tony Zhao: "In order for a live demo to work, it's really stressing all parts of the system. From the bottom is the reliability of the hardware, then the software system, then the model. Because we're a full-stack company, we're able to do excellent work across all of them. We have these robots running pretty much every day, and it just keeps getting better as they practice." brexton: "So kids can bump into Memo, people can drop clothes, and it will correct itself just on the fly." Tony Zhao: "Act Two is the first robotic policy that can handle the diversity of homes in a way that is this reliable. You can see as I'm twisting it, if I twist it in again, it will keep doing it. So Memo has infinite patience." Sunday
5:32

Sunday Robotics' Tony Zhao runs a fully autonomous robot demo live and unscripted, and it corrects itself every time you mess with it: brexton: "I want to confirm for the audience right now that this is fully autonomous. There's no teleoperation, no one behind the curtain, and this is real time. And we did not prepare this demo before. Why do you think seeing real-time demos is so rare for robotics? What the audience is seeing right now is not that common." Tony Zhao: "In order for a live demo to work, it's really stressing all parts of the system. From the bottom is the reliability of the hardware, then the software system, then the model. Because we're a full-stack company, we're able to do excellent work across all of them. We have these robots running pretty much every day, and it just keeps getting better as they practice." brexton: "So kids can bump into Memo, people can drop clothes, and it will correct itself just on the fly." Tony Zhao: "Act Two is the first robotic policy that can handle the diversity of homes in a way that is this reliable. You can see as I'm twisting it, if I twist it in again, it will keep doing it. So Memo has infinite patience." Sunday

MTS

134,079 Aufrufe • vor 1 Monat