Sharpa

🚀 We're coming to RSS 2026 in Sydney! Come to our booth near the main entrance to see how our dexterity-first technologies enable robust real-world productivity. We're looking forward to meeting researchers, engineers and founders from across the AI robotics community. 📍 Main Entrance 📅 July 13–16 📍 ICC – Sydney Stop by and say hello—we'd love to connect. #Sharpa #RSS2026 #Robotics #EmbodiedAI #DexterousManipulation #TactileSensing #HumanoidRobotics

NVIDIA's new CHORD framework teaches robot hands to manipulate objects by matching how human contact moves an object. Learned from human demos and transferred to real dexterous hands. Look at the high success rates: 82.12% success across 1,831 contact-rich tasks; 90.77% success on whole-body manipulation. With policies deployed from simulation. That's the #SharpaWave in the loop :) Project: nvidia-isaac.github.io/video_to_data/… Paper: nvidia-isaac.github.io/video_to_data/… #PhysicalAI #DexterousManipulation #RobotLearning #NVIDIA #EmbodiedAI #Robotics #Dexteroushands

@Dr_YanChang Great to see such high success rates when the policy is transferred to real robots! And proud to see Wave in there 🙌

How can robots learn dexterous manipulation from human demonstrations at scale? Excited to share CHORD: Learning Dexterous Manipulation Using Contact Wrench Guidance From Human Demonstration. CHORD learns from human demos by focusing not only on where contact happens, but how that contact moves the object through force and torque guidance. This unified contact-wrench representation carries human manipulation skills across diverse behaviors, long-horizon tasks, whole-body embodiments, and real-world hardware. We evaluated CHORD on large-scale, long-horizon, contact-rich tasks paired with human demonstrations, spanning rigid, articulated, and multi-object manipulation. At scale: * 82.12% average success across 1,831 tasks * 90.77% whole-body manipulation success * 4,739 sim-ready dexterous manipulation benchmark * Transfer to real dexterous hands Project page: nvidia-isaac.github.io/video_to_data/… Tech report: nvidia-isaac.github.io/video_to_data/… Code will be released soon as part of Video to Data repo github.com/nvidia-isaac/v…, our end-to-end pipeline for converting human demonstration videos into simulation-ready assets and physics-grounded robot training data. Huge thanks to amazing contributors: @zhu_xinghao , Zixi Liu, Shalin Jain, Chenran Li, Milad Noori, Huihua Zhao, John Welsh, @michaelv03, Wei Liu, @TingwuWang , Xingye (Dennis) Da, @zhengyiluo, Vishal Kulkarni, @sNaema, @yukez, @DrJimFan, @bowenwen_me, @danfei_xu, @SohaPouya, @Dr_YanChang. #Robotics #PhysicalAI #DexterousManipulation #RobotLearning #NVIDIA

@BotNewsAI @kushalk_ @tylerlum23 @leto__jean @KarenJLiu Thanks ;)

@SharpaRobotics @kushalk_ @tylerlum23 @leto__jean @KarenJLiu Sharpa is art

Before a robot can perfect assembly, it needs to learn to play. The team behind SimToolReal @kushalk_ @tylerlum23 @leto__jean @KarenJLiu 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. training from scratch, and zero-shot sim-to-real down to 0.5mm clearance. Peg insertion, screwing, multi-part assembly: all running at 60Hz, real speed, real hardware. And when a grasp slips, the policy doesn't stop, it recovers and keeps going. The Sharpa Wave responded present again ;) Project: play2perfect.github.io #Robotics #SharpaWave #Sharpa #EmbodiedAI #DexterousManipulation #RobotLearning

Project Page: ftp1-policy.github.io Arxiv: arxiv.org/abs/2606.13102 Code: github.com/michaelyuancb/… Model: huggingface.co/MJJJJ1064/ftp1… Dataset: modelscope.cn/datasets/Eurek…

What if the next leap in robot manipulation comes from touch, not just vision? To get there, foundation models need to understand tactile feedback the way they understand images and language. And tactile policies cannot be locked to specific hardware (that makes real-world deployments & maintenance quite complicated). FTP-1 solves both. One of the 1st foundation model for touch. 21 sensors. ~3,000 hours of data. Transfers to hardware it has never seen before. +17% on known hardware. +31% on never-seen hardware. We're proud this research led by @michaelyuancb ran on #SharpaNorth, #SharpaWave hands, and our DTC sensors. Special thanks to the teams at @Tsinghua_Uni , @UCBerkeley , @ETH , and @sjtu1896. Project page: ftp1-policy.github.io

Introduce FTP-1, the first Generalist Foundation Tactile Policy. Enjoy FTP-1 on any tactile sensors and embodiments in your labs🤗! Pretrained on 3000 hours tactile data and 21 sensors🤖, FTP-1 learns general tactile knowledge, that can even transfer to unseen sensors 🚀! FTP-1 is distributed and evaluated by 5 global institutions, including Sharpa, UC Berkeley, Tsinghua, ETH Zurich, Shanghai Jiaotong University. We fully open-source all data and checkpoints for community usage🤗. Check the blog for more details😃! ftp1-policy.github.io

@DuinoDu actual robot

@SharpaRobotics Is this a simulation or the actual robot? It feels very smooth.

North recently tried ironing. This week: shirt folding. 👕 Unlike ironing, folding is a skill even the most reluctant of our engineers have mastered. Notice how smooth the hand movements are. Clothes deserve some respect. 😉

@notmahi Thanks so much, Mahi! Our team is truly inspired by the groundbreaking research coming out of your lab. The results have been remarkable, and we're excited to see what comes next.

One of the underappreciated factor in learning dexterity from humans is the hardware – imitating humans the way we did would be much more difficult if we didn't have a roughly human sized hand. Thank you @SharpaRobotics for your support!

1 of 3: T-Rex, a new model and dataset for tactile-reactive dexterous manipulation. Tactile is a key modality for precise dynamic control:

The dataset was collected using two Sharpa Wave hands, using our high-resolution Dynamic Tactile Arrays on each fingertip to capture the dynamic touch signals needed for contact-rich manipulation. T-Rex proves that anthropomorphic hardware and tactile-reactive software are co-dependent :) Amazing work @Dantong_Niu ! P.S. The Wave's Isaac Sim URDF/USD assets and tactile parameters are available here if you wanna take a closer look: sharpa.com/pages/downloads

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 @DrJimFan, @drfeifei, @JitendraMalikCV, @pabbeel @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: tactile-rex.github.io

@Dantong_Niu worth spending some time reading, for anyone interested in mapping hand primitive to specific tasks and objects!

4/N: Why motor primitives? A key design choice behind the T-Rex Dataset is to prioritize motor primitives rather than collecting large amounts of task-specific demonstrations. Instead of focusing on a small set of benchmark tasks, we organize data around 22 elementary manipulation primitives: pick, place, slide, insert, extract, wipe, fold, squeeze, peel, wrap, ... Combined with 200+ everyday objects, this produces diverse contact-rich interactions while remaining highly data-efficient.

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: tactile-rex.github.io 📄 Paper: arxiv.org/abs/2606.17055 💻 Code: github.com/ZhuoyangLiu200… 🤗 Dataset: huggingface.co/datasets/zekai… 🧵 Thread ↓

@Dantong_Niu great dataset!

🦖 T-Rex Dataset To enable tactile-grounded mid-training, we collect: • 100 hours of tactile-synchronized dexterous manipulation • 7,700+ trajectories • 22 motor primitives • 200+ everyday objects Instead of collecting narrow task demonstrations, we organize data around compositional motor primitives and object interactions. This provides broad coverage of contact-rich manipulation while remaining data efficient. 50 hours of the dataset is open-sourced!

@Dantong_Niu Amazing work from @Dantong_Niu and team! A foundational work on demonstrating the value of tactile signals for dexterous manipulation policies. The work on hand movement primitive classification is remarkable, do take a look at Dantong's methodology! 👏