Wow. This is huge.
Researchers from UC Berkeley, NVIDIA and Stanford introduced T-Rex, a framework that combines vision, language and touch so robots can react to physical contact in real time instead of relying on vision alone.
This system is trained on a 100-hour tactile-synchronized dataset spanning 200+ everyday objects and 22 motor primitives, teaching robots how to feel and manipulate the physical world.
Across 12 contact-rich manipulation tasks, T-Rex achieved over 30% higher average success rates than the strongest baseline models.
Perhaps the "GPT-3 moment" for humanoid robots lies within vast amounts of human video data.
UC Berkeley (Abbeel + Malik group) just released Do as I Do: a pipeline that takes monocular RGB human videos (ego + exo)->reconstructs hand-object interactions->retargets them to multi-fingered robotic hands, generating executable robot trajectories.
It pushes success rates on real-world clips from ~25% to 71%, outperforming prior SOTA.
This kind of human video->robot-complete data approach feels increasingly important for closing the data flywheel in humanoid/embodied AI.
Researchers from @UCBerkeley, @nvidia, and @Stanford introduce T-Rex, a framework that unifies vision, language, and tactile sensing so robots can respond to physical contact in real time rather than relying on vision alone.
The foundation is a 100-hour tactile-synchronized teleoperation dataset spanning 200+ everyday objects and 22 motor primitives. During data collection, researchers wore @ManusMeta gloves to capture precise finger motion, which was then retargeted onto @SharpaRobotics Wave dexterous hands for bimanual teleoperation.
Code: https://t.co/SyeiGeEXoP
Excited to share T-Rex: Tactile-Reactive Dexterous Manipulation 🦖🤖
Touch is fundamental to human dexterity, yet most Vision-Language-Action (VLA) models either ignore tactile feedback or lack the ability to react to high-frequency contact signals.
In this work, we tackle both the data and architectural challenges of tactile-reactive dexterous manipulation.
🦖 A 100-hour tactile-synchronized dexterous manipulation dataset with 7,700+ trajectories, 22 motor primitives, and 200+ everyday objects.
🦖 A tactile-reactive MoT architecture with spatial-temporal tactile encoding and asynchronous high-frequency tactile refinement.
🦖 A scalable training recipe combining 22,889 hours of human egocentric pretraining with tactile-grounded robot mid-training.
Across 12 real-world contact-rich manipulation tasks, T-Rex achieves over 30% higher average success rate than the strongest baseline.
We are fully open-sourcing the dataset, models, teleoperation stack, training code, and inference pipeline.
🌐 Project: https://t.co/AiHKRR8YXU
📄 Paper: https://t.co/mXY2UNLlqc
💻 Code: https://t.co/7skCxUtwKC
🤗 Dataset: https://t.co/uNwW8dcRZL
🧵 Thread ↓
This giant free dataset could make helper robots way smarter, way faster:
An open-source robotics stack from Berkeley AI researchers featuring the largest teleoperation dataset released to date with over 3,500 hours of bimanual manipulation data across 200 tasks.
The video showcases autonomous bimanual robot performance on dexterous tasks including box folding, Lego sorting, AirPod insertion, t-shirt folding, backpack packing, and box unlocking using learned policies.
Sim-to-real correlations, training insights like flow loss predicting real-world success, lightweight infrastructure for DAgger interventions
Thank you for sharing, @ritvik_singh9, and everyone else who contributed to this!
Links to the paper plus dataset at https://t.co/4cNtRw02mf under permissive licensing.
——-
Weekly robotics and AI insights.
Subscribe free: https://t.co/9Nm01QUcw3
When a robot learns to react to touch, where does the training data come from?
Researchers from @UCBerkeley, @nvidia, and @Stanford introduce T-Rex, a framework that unifies vision, language, and tactile sensing so robots can respond to physical contact in real time rather than relying on vision alone.
On contact-rich tasks like inserting a card, turning a key, and handling deformable objects, it outperforms the strongest baseline by more than 30% across 12 real-world tasks.
The foundation is a 100-hour tactile-synchronized teleoperation dataset spanning 200+ everyday objects and 22 motor primitives. During data collection, researchers wore @ManusMeta gloves to capture precise finger motion, which was then retargeted onto @SharpaRobotics Wave dexterous hands for bimanual teleoperation.
Learn more: https://t.co/Kz85DsTxR0
We can use videos from the internet to teach robots!
Do as I Do, from
@bhawna_paliwal_ , @HarithejaE , and Willian Liang at UC Berkeley — advised by @pabbeel, @notmahi, and @JitendraMalikCV, used an algorithm that reconstructs hand-object interactions from monocular RGB video and retargets them into real, executable trajectories for multi-fingered dexterous hands. Just using "low quality" video footage of humans doing tasks. No sensors.
The Sharpa Wave robot hand being anthropomorphic, it matches human kinematics. Not only that works, but at fast speed, too!
Congrats to the team, that's super exciting!
Project: https://t.co/khlcaUYCEp
#Robotics #SharpaWave #Sharpa #EmbodiedAI #DexterousManipulation #RobotLearning
Hyundai and Boston Dynamics’ Ghost Rabona work with Atlas is a strong example of how far motion learning in humanoid robots has progressed.
In this work, Atlas does not perform the movement through manually programmed commands from scratch. Instead, it learns from reference motions based on human demonstrations, combined with reinforcement learning and simulation-based training.
First, high-quality kinematic data is collected from a football player through motion capture. This human motion is then adapted to Atlas’ own body structure and joint constraints. In simulation, the robot learns to solve balance, motor control, and timing challenges through thousands of parallel trials.
The Ghost Rabona is particularly difficult because the robot needs to execute a fast fake movement, control its center of mass, maintain balance on one leg, and complete a powerful kick through full-body coordination.
It seems that we may not be too far from seeing longer and more complete football-specific humanoid robot skill videos. What do you think?
🎥 via Hyundai and Boston Dynamics
ℹ️ This content is shared for informational purposes only. All intellectual property rights belong to their respective owners.
#HumanoidRobots #Hyundai #BostonDynamics #AtlasRobot #PhysicalAI #ReinforcementLearning #SimToReal #Robotics
Sprout is a compact humanoid robot developed by Fauna Robotics for research, education, and human-robot interaction.
Standing around 3.5 feet tall and weighing roughly 50 pounds, Sprout is designed to operate safely in spaces shared with people.
Watch this humanoid robot crush an agility course then casually grab zongzi like a pro.
From ramps and hurdles to precise picking SDIC Intelligence's bot nails dynamic balance jumps and fine motor skills.
Perfect timing for Dragon Boat Festival!
China's humanoid push is accelerating fast. What's next?
#HumanoidRobot #AI #DragonBoatFestival #Robotics
(Video from SDIC Intelligence / Meiya Pico demo)
LimX Dynamics has introduced Luna, a 1.6-meter-tall humanoid robot built for entertainment and interactive performances.
It can dance, perform gestures, and interact with people using smooth movements and digital facial expressions.
Luna also features AI-powered motion generation, zero-code programming, and multiple safety systems for public use.
Turns out you can train humanoid hands without any robot data.
The idea in HUG is quite simple: (a) collect human data with smart glasses, (b) train a human manipulation model, (c) retarget to multi-fingered robot hands.
Meet Sprout, a small humanoid robot developed by Fauna Robotics for research, education, and social interaction.
It can walk, climb stairs, and safely interact with people with its lightweight design and soft exterior.
Fauna Robotics was acquired by Amazon, and Sprout continues to serve as a platform for robotics research and development.
This low-cost exoskeleton could make robot training 10x faster:
A low-cost exoskeleton that provides real-time haptic torque feedback, allowing a human teleoperator to feel and control a robot’s forces during complex manipulation tasks while capturing data for autonomous policies.
Demonstrations include a blindfolded user unsheathing a sword via the mirrored robot arm and autonomous robots performing force-sensitive tasks like retrieving a drink from a fridge side holder or navigating tight spaces in a kitchen.
The system uses bidirectional kinematics and dynamics retargeting for compatibility with different robots.
Enabling rapid data collection, overnight training, and next-day autonomy for whole-body, occluded, and long-horizon mobile manipulation.
Thanks for sharing this great work, @litian_liang!
📌 https://t.co/S3XJKXI3NF
——
Weekly robotics and AI insights.
Subscribe free: https://t.co/9Nm01QUcw3
Shanghai-based robotics company AGIBOT has updated its Lingxi X2 humanoid robot with smarter motion capabilities.
The robot can detect moving objects around it, predict their path, and quickly move out of the way.
Lingxi X2 can also climb stairs, keep its balance on uneven surfaces, and move more smoothly.