Skip to content

XTac UMI G1

Xense XTac UMI G1 is a wearable data-collection gripper system for building embodied manipulation datasets. The leader gripper integrates tri-color visuotactile sensors, a wrist fisheye camera and a high-precision pose sensor, capturing vision, touch, pose, gripper state and IMU data in sync; the follower gripper mounts on the robot's end effector to execute or replay motions. The system serves imitation learning, VLA / VTLA model training and tactile world model research.

XTac UMI G1 product photo

System components

Part Role
Leader grippers ×2 (left / right) Handheld demonstration; each carries two visuotactile sensors, a wrist fisheye camera, an opening encoder and an IMU, see Gripper connection and serial numbers
Pico4 Ultra Enterprise trackers ×2 Mounted on top of the leader grippers, giving their 6-DoF pose
Pico4 Ultra Enterprise headset Tracks the trackers and sets the world frame; can also record a first-person stereo view, see Pico4 headset and tracker setup
Collection unit The backpack on the Backpack Kit, your own collection PC on the Developer Kit, see Product line and configuration comparison
Follower grippers (optional) Mounted on a robot to execute or replay motions, see Follower gripper overview

Specifications are in Technical specifications.

Data flow

The sensor side is identical in both editions; they differ only from the collection unit onward. Hover over (or tap) any block to see where its data comes from and where it goes.

Backpack Kit

Recording is controlled by gripper button gestures; the gesture table is in Buttons.

Developer Kit

Key differences between the editions

Backpack Kit Developer Kit
Preview Browser console, a tablet is enough Rerun window
Recording control Mainly the gripper buttons; the console works too lerobot-record command line
On disk One MCAP raw recording per episode LeRobotDataset v3 directly
Data out Export per task as LeRobotDataset v3 / MCAP, upload to ModelScope in one click Push to the Hugging Face Hub
Open source No Yes (Apache-2.0)

Both editions record poses in the same coordinate frame, see Coordinate frames. The full comparison, box contents and wiring are in Product line and configuration comparison.

Next steps