Jialei Huang retweeted
Seems like a great tool for high quality, scalable dexterous data collection
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: x2robot.com/en/pages/twindex
#TwinDEX #Robotics #EmbodiedAI #DexterousManipulation
Proud to share what our team and I have been building over the past few months — TwinDEX: a twinned system with two co-designed three-finger dexterous manipulators, one wearable for data collection, one for robot deployment. Identical kinematics, identical contact surfaces, identical sensors — a scalable path from robot-free demonstrations to real-world dexterous manipulation.
Why twin? Because when collection and deployment are truly the same, robot-free data works just as well as on-robot teleoperation — no retargeting, no domain gap, no fine-tuning.
Trained from scratch on a few hundred wearable demos with zero on-robot data, TwinDEX completed a 24-step chemistry experiment: cap twisting, pipette operation, glass-rod guided pouring, bimanual coordination — each demanding millimeter-level precision and stable force control.
The nail-it phase is done. Scale-it is underway.
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: x2robot.com/en/pages/twindex
#TwinDEX #Robotics #EmbodiedAI #DexterousManipulation
Jialei Huang retweeted
𝑪𝒐-𝒕𝒓𝒂𝒊𝒏𝒊𝒏𝒈 is a promising way to scale Large Behavior Models (LBMs) beyond robot data, yet the data and training recipe are far from settled. 🤔
We present a large-scale empirical study leveraging 4,000h of robot/human data and 50M vision-language samples, evaluating 89 policies across 58,000 simulation rollouts and 2,835 real-world trials. 🤖📊
co-training-lbm.github.io/
Work done during my internship at @ToyotaResearch.
Jialei Huang retweeted
Having humanoid to do repetitive jobs is the dream that I have for a long time. Now we are making a steady step towards it!
🏭 Humanoid robots have reached a new stage: scaled production.
🎉Spirit AI’s Moz1 humanoid robots are now operating at scale on a CATL battery pack line—the first-ever deployment of its kind.
🤖 It solves real problems: automating high-voltage, high-precision plugging with >99% success rate, ensuring safety and consistency.
🧠 Powered by our end-to-end Spirit VLA model, Moz1 adapts in real-time to variations—this is Embodied Intelligence moving from lab to factory floor.
🚀A major step for intelligent, flexible automation.
#EmbodiedAI #Tech #Robotics #FutureOfWork
Teaching robots sub-millimeter precision through spatially-grounded touch
We introduce SaTA which anchors tactile to hand kinematics, enabling learning-based policies to match model-based precision.
arxiv.org/abs/2510.14647
(2/3)Why it matters: At contact, vision fails. Traditional learning treats tactile as abstract features—knows "something touched" but not "how to adjust."
SaTA preserves tactile richness AND spatial context → sub-mm precision
What if a robot could reason through failures like "the board isn't clean... I need more force"—and fix it autonomously? Excited to share "Tactile-VLA: unlocking VLMs' latent physical knowledge for contact-rich manipulation through tactile sensing" in jialeihuang.github.io/tactil…
2/ Leveraging VLM common sense Without explicit force instructions, the robot automatically applies firm grip to heavy iron ball, gentle touch to fragile pitaya—objects never seen. The VLM's prior knowledge about object properties transfers directly to appropriate contact forces.
1
3
422








