Quickstart

From power-on to your first recorded episode. Before you start, make sure the following five items are complete:

• Hardware connected → Leader gripper connection and use
• Environment installed and active (mamba activate xense-taccap or the Docker container) → Installation
• Collection PC configured: brltty removed, serial permissions, ModemManager kept off the ports (miss one and the gripper will not connect) → Host setup · Required on first use
• Every leader gripper calibrated (an uncalibrated leader is refused) → Gripper calibration (zero + travel)
• Repo, SDK and gripper firmware all up to date → You must upgrade to the latest versions

1. Power on and connect

1. Plug in the gripper USB.
2. Connect the headset to the collection PC by cable and turn off the collection PC's Wi-Fi.
3. Turn on the headset and short-press the tracker power button until the LED is solid blue.
4. Start the XenseVR PC Service: /opt/apps/roboticsservice/runService.sh
5. Facing the same direction as the robot, open XTac UMI XR, choose "USB Network" as the connection and tap "Connect"; Status changes to "Connected".

The gripper connects to the collection PC over USB Type-C

XTac UMI XR Status changes to "Connected"

Three common mistakes

• Start the PC Service before opening the XR app, or it cannot connect.
• With a wired headset the collection PC's Wi-Fi must be off, or tracking is unstable. See Network connection.
• Do not restart the XR app during collection: it resets the world origin, so poses within one dataset stop lining up.

Power off in reverse: stop recording and let the current episode save, quit the XR app, stop the PC Service, then unplug, in the order given in Power and connection requirements.

2. Self-check

python -c "from xense.taccap import scan_grippers
for g in scan_grippers(): print(g.side.name, g.role.name, repr(g.firmware_sn))"

One line per gripper, role is Leader or Follower, and the serial is not empty. If not, see Troubleshooting.

3. Preview check

② With wrist pose③ Full setup

lerobot-teleoperate \
    --robot.type=bi_taccap_gripper \
    --robot.id=0 \
    --fps=30 \
    --display_data=true

lerobot-teleoperate \
    --robot.type=xtac_umi_g1 \
    --robot.id=0 \
    --fps=30 \
    --display_data=true

Move and open/close the grippers, check that every image, tactile stream and pose updates, then press Ctrl+C.

The tiers are chosen with --robot.type; record with the same tier you previewed:

Tier	How	Data included
① Grippers only	--robot.type=bi_taccap_gripper --robot.enable_tracker=false	Tactile, wrist cameras, gripper opening; no PC Service needed
② With wrist pose	--robot.type=bi_taccap_gripper	Adds the gripper pose tcp.*
③ Full setup	--robot.type=xtac_umi_g1	Adds headset stereo images and head pose

Keep the trackers in the headset's view before starting; tracking is lost if they are occluded.

4. Full recording

② With wrist pose③ Full setup

lerobot-record \
    --robot.type=bi_taccap_gripper \
    --robot.id=0 \
    --dataset.repo_id=<your_org>/<dataset_name> \
    --dataset.single_task='Pick up the object' \
    --dataset.num_episodes=1 \
    --dataset.fps=30 \
    --dataset.episode_time_s=120 \
    --dataset.reset_time_s=60 \
    --dataset.push_to_hub=false

Each frame records 6 image streams (four visuotactile, both wrist cameras), plus gripper opening and gripper pose.

lerobot-record \
    --robot.type=xtac_umi_g1 \
    --robot.id=0 \
    --dataset.repo_id=<your_org>/<dataset_name> \
    --dataset.single_task='Pick up the object' \
    --dataset.num_episodes=1 \
    --dataset.fps=30 \
    --dataset.episode_time_s=120 \
    --dataset.reset_time_s=60 \
    --dataset.push_to_hub=false

Each frame records 8 image streams, adding the headset stereo pair, plus the head pose:

The eight streams and their keys in the dataset

• --robot.id is required: the station number as a plain number (0, 1, …), one per setup.
• Keep --robot.type (and whether you add --robot.enable_tracker=false) the same as in the preview; for grippers only, add that switch to the ② command.
• Single gripper: --robot.type=taccap_gripper; with both grippers plugged in, add --robot.side=left or right.

All parameters: Data collection · Parameters.

5. Check and upload

Check the dataset:

lerobot-check-dataset --repo-id <your_org>/<dataset_name>

Upload to the Hugging Face Hub when needed:

lerobot-push-dataset-to-hub \
    --repo-id <your_org>/<dataset_name> \
    --dataset-path ~/.cache/huggingface/lerobot/<your_org>/<dataset_name> \
    --upload-large-folder

Dataset layout and fields: Dataset.