Wearable Data Collection

Having a person wear cameras, gloves, and similar devices while doing their normal job, recording their motion as robot training data.

Wearable data collection is one of the main forms of embodiment-free data collection (collecting data without needing an actual robot). A collector wears a head- or chest-mounted camera and a wrist-mounted camera, often paired with a data glove, an IMU (inertial measurement unit, measuring acceleration and angular velocity), or tactile sensors, and does their normal job in real settings like factories, supermarkets, or homes, while the devices synchronously record first-person video, depth, and hand and body pose. Compared with teleoperation, it doesn't require moving a robot to the site, is cheaper, produces more natural motion, and scales up easily; the tradeoff is that a human hand's structure differs from a robot hand's, so the motion needs retargeting or additional modeling before a robot can use it. Compared with a handheld gripper like UMI, it captures richer hand motion, but the correspondence to a robot's end effector is weaker. The engineering challenges are multi-sensor time synchronization, calibration, recovering from occlusion, and unifying coordinate frames.

ExampleTARS Robotics has collectors wear its own capture kit while working in factories, supermarkets, hotels, and other settings, recording over 1,000 hours of human manipulation data, organized into the WIYH dataset.

Related

Robot-free (Embodiment-free) Data Collection · Handheld Gripper Data Collection · Egocentric Video · Data Glove · Motion Retargeting · Human Video Data

Sources

无本体数据采集技术演进：从 UMI、可穿戴采集到规模化交付（新浪科技，2026-09） (Chinese)
arXiv 2512.24310: World In Your Hands

As of

2026-09