Genrobot releases DAS Dex hand capture
What happened
Genrobot unveiled DAS Dex, a 23‑degree‑of‑freedom hand capture system that fuses tactile and vision data with millimetre precision and sub‑millisecond synchronization, aiming to improve embodied‑AI training for dexterous manipulation. The system targets high‑fidelity hand data collection for embodied learning and control (x.com).
Why it matters
Genrobot has unveiled DAS Dex, a wearable hand‑capture system that records human hand motion, touch, and wrist vision together at very high fidelity. (genrobot.ai) The device straps on like a slim exoskeleton and measures 23 degrees of freedom to reproduce the full range of human finger and palm motion. (genrobot.ai) Sensors include magnetic encoders for joint angles, high‑resolution tactile arrays on the fingers, and a wrist‑mounted camera that captures the hand’s view; the company says fingertip position is accurate to about one millimetre and joint angles to 0.02°. (genrobot.ai) DAS Dex streams synchronized multimodal data — fingertip trajectories, per‑finger 3‑D touch maps, joint angles, and wrist video — at hundreds of frames per second with sub‑millisecond timing between channels. (genrobot.ai) Those timing and spatial numbers matter because training a robot to manipulate objects requires knowing not just where a finger was, but exactly when and how it contacted a surface; mixed sensors that are even slightly out of sync can make contact events ambiguous. (genrobot.ai) Genrobot positions DAS Dex as part of a larger data‑first playbook for “embodied intelligence”: collecting large volumes of natural human interactions so models can learn how bodies move and touch things in the real world. (genrobot.ai) The company already offers complementary devices — for example a head‑mounted DAS Ego system for first‑person video and wider scene capture — and pitches DAS Dex as the hand half of a full “head + hand” data stack. (genrobot.ai) In practice, teams record a human demonstrator performing tasks while DAS Dex reconstructs the hand’s pose and tactile contacts; that recorded trace can be replayed to analyze grasps or used as supervised data to train controllers, imitation learners, or simulation environments. (genrobot.ai) High‑quality human hand data is scarce relative to the demand for dexterous manipulation, so a device that reliably captures touch and motion together can shorten the loop from demonstration to deployable robot behavior. (dexcanvas.github.io) For researchers and engineers, the system also exports common formats and provides SDKs and ROS support so captured sequences can be fed into perception pipelines, grasp planners, or physics‑aware training frameworks. (docs.genrobot.ai) On the hardware side DAS Dex is compact and battery powered, with on‑device storage and Wi‑Fi streaming; Genrobot lists a roughly three‑hour runtime, 64 GB storage, and output at about 200 Hz. (genrobot.ai) The device is aimed at labs and companies building embodied‑AI models and dexterous manipulators, not hobbyists — the product pages emphasize integration with cloud tooling and large‑scale data pipelines rather than a low price point. (genrobot.ai) For a student or engineer thinking about careers in robotics, DAS Dex signals the industry’s heavy emphasis on data quality: better sensors and synchronization reduce downstream modeling work and make learned controllers more reliable when they touch real objects. (genrobot.ai) Genrobot’s product page lists the complete technical sheet and contact route for trials or purchase. (genrobot.ai)
Key numbers
- Genrobot unveiled DAS Dex, a 23‑degree‑of‑freedom hand capture system that fuses tactile and vision data with millimetre precision and sub‑millisecond synchronization, aiming to improve embodied‑AI training for dexterous manipulation.
- (genrobot.ai) The device straps on like a slim exoskeleton and measures 23 degrees of freedom to reproduce the full range of human finger and palm motion.
- (genrobot.ai) DAS Dex streams synchronized multimodal data — fingertip trajectories, per‑finger 3‑D touch maps, joint angles, and wrist video — at hundreds of frames per second with sub‑millisecond timing between channels.
- (docs.genrobot.ai) On the hardware side DAS Dex is compact and battery powered, with on‑device storage and Wi‑Fi streaming; Genrobot lists a roughly three‑hour runtime, 64 GB storage, and output at about 200 Hz.
What happens next
- The system targets high‑fidelity hand data collection for embodied learning and control (x.com).
Quick answers
What happened in Genrobot releases DAS Dex hand capture?
Genrobot unveiled DAS Dex, a 23‑degree‑of‑freedom hand capture system that fuses tactile and vision data with millimetre precision and sub‑millisecond synchronization, aiming to improve embodied‑AI training for dexterous manipulation. The system targets high‑fidelity hand data collection for embodied learning and control (x.com).
Why does Genrobot releases DAS Dex hand capture matter?
Genrobot has unveiled DAS Dex, a wearable hand‑capture system that records human hand motion, touch, and wrist vision together at very high fidelity. (genrobot.ai) The device straps on like a slim exoskeleton and measures 23 degrees of freedom to reproduce the full range of human finger and palm motion. (genrobot.ai) Sensors include magnetic encoders for joint angles, high‑resolution tactile arrays on the fingers, and a wrist‑mounted camera that captures the hand’s view; the company says fingertip position is accurate to about one millimetre and joint angles to 0.02°. (genrobot.ai) DAS Dex streams synchronized multimodal data — fingertip trajectories, per‑finger 3‑D touch maps, joint angles, and wrist video — at hundreds of frames per second with sub‑millisecond timing between channels. (genrobot.ai) Those timing and spatial numbers matter because training a robot to manipulate objects requires knowing not just where a finger was, but exactly when and how it contacted a surface; mixed sensors that are even slightly out of sync can make contact events ambiguous. (genrobot.ai) Genrobot positions DAS Dex as part of a larger data‑first playbook for “embodied intelligence”: collecting large volumes of natural human interactions so models can learn how bodies move and touch things in the real world. (genrobot.ai) The company already offers complementary devices — for example a head‑mounted DAS Ego system for first‑person video and wider scene capture — and pitches DAS Dex as the hand half of a full “head + hand” data stack. (genrobot.ai) In practice, teams record a human demonstrator performing tasks while DAS Dex reconstructs the hand’s pose and tactile contacts; that recorded trace can be replayed to analyze grasps or used as supervised data to train controllers, imitation learners, or simulation environments. (genrobot.ai) High‑quality human hand data is scarce relative to the demand for dexterous manipulation, so a device that reliably captures touch and motion together can shorten the loop from demonstration to deployable robot behavior. (dexcanvas.github.io) For researchers and engineers, the system also exports common formats and provides SDKs and ROS support so captured sequences can be fed into perception pipelines, grasp planners, or physics‑aware training frameworks. (docs.genrobot.ai) On the hardware side DAS Dex is compact and battery powered, with on‑device storage and Wi‑Fi streaming; Genrobot lists a roughly three‑hour runtime, 64 GB storage, and output at about 200 Hz. (genrobot.ai) The device is aimed at labs and companies building embodied‑AI models and dexterous manipulators, not hobbyists — the product pages emphasize integration with cloud tooling and large‑scale data pipelines rather than a low price point. (genrobot.ai) For a student or engineer thinking about careers in robotics, DAS Dex signals the industry’s heavy emphasis on data quality: better sensors and synchronization reduce downstream modeling work and make learned controllers more reliable when they touch real objects. (genrobot.ai) Genrobot’s product page lists the complete technical sheet and contact route for trials or purchase. (genrobot.ai)
