MANUS™

Our research shows that cat videos are still more popular than haptic data glove videos. We were baffled by this result too. So we decided to combine both. Check it out and feel the purr-fection.

@nvidia 's DreamDojo world model pretrains on roughly 44,000 hours of egocentric human video, using latent actions as proxy labels for dexterous robot learning. To validate this approach, researchers used MANUS gloves paired with the Vive Ultimate Tracker to capture precise hand poses during manipulation tasks. These were retargeted into ground-truth actions for the @FourierRobots GR-1 humanoid, creating a high-precision baseline for comparing latent action conditioning against ground-truth motion capture. Read more: manus-meta.com/use-cases/manu…

Robot learning depends on the quality of the data behind it. In the @ABBRobotics and @PSYONICinc collaboration, MANUS gloves captured natural hand movements as human demonstration data, while @HaplyRobotics handled wrist tracking and force feedback. This kind of workflow reflects a broader shift. As imitation learning and teleoperation become more common, capturing human motion with fidelity is becoming a core part of building useful training datasets. Read more: manus-meta.com/use-cases/abb-…

Most people think of AI as LLMs like ChatGPT or Claude. But language intelligence is only part of it. For a robot to manipulate the world as dexterously as a human, and to relieve people of repetitive, dangerous, and strenuous work, it needs action data to learn from. Unlike language, that data can't be scraped from the internet. That's why leading research labs and robotics companies building dexterous systems use MANUS gloves to collect teleoperation and human demonstration data, in three main ways: 1️⃣Egocentric data collection. High-fidelity kinematic hand motion, occlusion free and drift free, for foundation models that generalize across embodiments. 2️⃣Real-world teleoperation. Low-latency finger tracking drives physical robot hands in real time, with optional haptic feedback on contact. 3️⃣Simulated teleoperation. Teleoperate a simulated robot inside @NVIDIARobotics Isaac Lab, reducing dependence on physical robots while preserving the policy quality needed for sim-to-real transfer. Learn more → manus-meta.com/robotics

TESOLLO teleoperates its DG-5F-M five-fingered hand with MANUS Metagloves Pro Haptic, in two setups: driven in simulation through @nvidia Isaac Sim, and the same control mapped to a real DG-5F-M. Tactile data streams back to the glove's haptic actuators, so the operator feels contact in either environment. The DG-5F-M is a 20-DoF anthropomorphic hand with four independently driven joints per finger, modeled on an adult hand at about 1.76 kg. It runs 250 Hz control with absolute encoders, handles pinch payloads up to 5 kg and envelope payloads up to 20 kg, communicates over Modbus RTU/TCP or Ethernet, and ships with ROS 2 support and a public development repository. @Tesollo_Inc

Teaching #Robots the Art of Skilled Handiwork Through #AI by @ManusMeta #Robotics #Engineering #Innovation #Technology

@Ronald_vanLoon Appreciate the shoutout Ronald!

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: tactile-rex.github.io

Bimanual manipulation from @roboterax, showing coordinated control of XHand 1 Pro dexterous hands, teleoperated using @ManusMeta haptic gloves. Each hand: 21 active DoF, fully direct-drive and all backdrivable. 18 distributed tactile sensors across the fingertips, finger pads, and palm. ±0.1 mm pitch-joint repeatability, 4 kHz current-loop force control, and an open development stack on Ubuntu with C++, Python, and ROS 2. Learn more about MANUS in your robotics pipeline → manus-meta.com/robotics

Watch Xynova's Flex 2 teleoperated in real time with a MANUS haptic glove. Xynova's Flex 2 is a 23-DoF biomimetic hand featuring ±0.1 mm repeatability and load-backdrivable actuation with force-position hybrid control down to 0.05 N. It integrates multimodal sensing for slip detection and compliant reflexes, and supports an open development ecosystem. Learn more: manus-meta.com/use-cases/how-…

A dexterous humanoid learned to sort by color, chain skills, and follow a placement order with no robot demonstrations of those tasks. @Stanford and @Meta's Ego-Pi co-trains human and robot data on a π0.5 VLA. 90%+ success. The setup: a Galaxea R1 Pro whose dexterous hand control comes from MANUS gloves capturing the operator's finger joint angles. The same rig collects the human demonstrations, so both embodiments share one finger-tracking stream. Aligning in joint-angle space avoids the robot-side IK that produces self-colliding poses on high-DOF hands. Co-training alone reached 92% on color sorting and 90% on packaging; skill chaining hit 93% once subtask prediction was added as an auxiliary loss, up from 27%. Learn more: egopipaper.github.io Paper: arxiv.org/abs/2606.08107

The future of Physical AI isn’t built by one type of person. At MANUS, people from different backgrounds come together to solve hard problems, move fast, and build technology used by robotics teams around the world. Here's a glimpse of the women helping shape it. Want to build with us? We're #hiring in Eindhoven, The Netherlands. 👇 manus-meta.com/careers

Excited to see NVIDIA bringing together the technologies that make scalable robot learning possible. Congrats!

What a week at ICRA 2026. A big thank you to everyone who stopped by and shared insights with us in Vienna. One of the highlights of #ICRA2026 was seeing MANUS™ gloves powering live demos across partner booths throughout the exhibition. Seeing so many teams rely on MANUS technology is a reflection of the trust we’ve built together over the years. As robotics continues to advance, high-quality human motion data remains a critical piece of the puzzle. We’re proud to support researchers, developers, and innovators pushing this forward. See you at the next one.

So cool to see @ManusMeta gloves running live demos across so many booths at #ICRA2026.

You can't walk the ICRA floor without running into MANUS gloves. Day 1 at #ICRA2026 was nonstop for us at booth 115, and across the hall MANUS gloves were powering live demos for many of the teams pushing dexterous manipulation, teleoperation, and embodied AI forward. There's a reason you keep seeing MANUS everywhere at ICRA. Try our gloves yourself, and the difference speaks for itself. See you at booth 115. @ieee_ras_icra #ICRA2026 #Robotics #EmbodiedAI #Teleoperation

@NVIDIAGTC @UnitreeRobotics @SharpaRobotics Read the #NVIDIAGTC Taipei announcement: nvidianews.nvidia.com/news/nvidia-op…

Great to see MANUS gloves featured in Jensen Huang's keynote at @NVIDIAGTC Taipei, teleoperating the Sharpa hands of the newly unveiled NVIDIA Isaac GR00T Reference Humanoid Robot. The reference design brings together a @UnitreeRobotics H2 Plus, @SharpaRobotics Wave dexterous hands, and @nvidia Jetson Thor running Isaac GR00T as the onboard brain. Our part sits in the data layer: capturing high-fidelity human hand motion, the demonstration data that teaches robots to manipulate with precision. MANUS is the data glove officially supported in NVIDIA Isaac Teleop. Congrats to our partners at @nvidia, @UnitreeRobotics, and @SharpaRobotics! Fine manipulation is one of the hardest problems in humanoid robotics, and it starts with good data. Heading to #ICRA2026 in Vienna, Austria? Meet the MANUS team at Booth 115 and feel the precision yourself.

NVIDIA announces the first open humanoid robot reference design built for robotics research. The NVIDIA Isaac GR00T Reference Humanoid Robot combines the @UnitreeRobotics H2 humanoid robot, @SharpaRobotics Wave five-fingered hands for dexterous manipulation, Jetson Thor onboard compute, and Isaac GR00T open software and models, giving researchers a full-stack platform from data capture to model deployment. Read the #NVIDIAGTC Taipei announcement: nvda.ws/4ef9VOr

@SharpaRobotics See you soon 🔥

We will be at ICRA in Vienna next week. We first revealed our first product, the Wave dexterous hand, at the same time during ICRA last year. Since then we've delivered the Wave to research labs across the world, built our full body robot North, and developed our tactile AI Craftnet that gives robots a reflex-like intelligence for manipulation tasks. Come to Booth #33 to see for yourself how we're advancing manipulation skills in robots!