What does this skill do, and when should you use it?
This skill records teleoperated episodes in an Isaac for Healthcare agentic environment. It supports keyboard, SO-ARM leader, and VR devices, with availability determined by the selected environment. The workflow launches the recorder through i4h-workflows and stores successful demonstrations in HDF5. It is a good fit for users with a configured workflow, a supported environment, and real teleoperation hardware who need reproducible demonstration data.
Resolves or prepares an i4h-workflows checkout and clones it from GitHub when needed; creates a run directory containing data and logs; launches workflows/agentic/arena/run.sh in teleoperation mode for a selected environment, device, and episode count; extracts device-specific keybindings from the startup log; writes recordings to an absolute HDF5 path; checks the output file and success count; and stops the run with stop.sh.
- A robotics researcher records successful demonstrations for scissor_pick_and_place using a keyboard or SO-ARM leader.
- An embodied-AI developer collects mobile manipulation data in locomanip_tray_pick_and_place with keyboard_23d.
- A research team saves multiple human-completed episodes as HDF5 for downstream dataset workflows.
- An operator verifies the visible GUI, recorder readiness, and startup control map before driving the simulation.
How do you install this skill?
- The default path may clone external code from GitHub and launch a robot/simulation workflow; confirm repository provenance, network reachability, working directory, and consent for external effects before execution.
- Recordings may contain camera frames, robot state, or other sensitive data, but the skill does not define classification, access control, retention, or secure-storage requirements.
- Without a human operator, only readiness verification is appropriate; do not claim the requested number of successful episodes without checking the HDF5, logs, and remaining processes.
- Shell / CLI
- Network access
- Local filesystem
Giti4h-workflows.venvIsaac Sim
Install the individual skill with the NVIDIA skills CLI: npx skills add nvidia/skills --skill i4h-workflow-dataset-teleop --yes. The skill also requires workflow setup and an existing .venv through i4h-workflow-setup; the supplied material does not document that setup procedure.
tmp="$(mktemp -d)"
git clone --depth 1 https://github.com/NVIDIA/skills.git "$tmp"
mkdir -p ~/.claude/skills
cp -R "$tmp/skills/i4h-workflow-dataset-teleop" ~/.claude/skills/
rm -rf "$tmp"Generated from the source repository and skill path; it copies only this skill's folder. If the author's install steps above differ, follow those first. To scope it to one project, replace ~/.claude/skills with that project's .claude/skills.
How do you use this skill?
Ensure I4H_WORKFLOWS points to a checkout containing workflows/agentic, or allow the skill to clone the workflow to ~/i4h-workflows when absent. Select an environment and supported device, such as scissor_pick_and_place with keyboard or so101_leader, then run arena/run.sh --env <env> --teleop --teleop-device <device> --episodes 3 --record-to <absolute-HDF5-path>. Wait for the log to print the complete device-specific keybinding table before the operator starts. Use B to start an episode, N to mark success and save, and R to discard and reset; F is reserved by Isaac Sim. Confirm that the HDF5 file exists and that the log reports a success count equal to the requested episode count.
What are this skill's strengths and limitations?
- Supports several teleoperation modes, including keyboard, SO-ARM leader, and VR.
- Defines a clear start, success, discard, and reset workflow.
- Includes procedures for surfacing keybindings and verifying the HDF5 output and success count.
- Supports multiple documented environments and reuse of an existing i4h-workflows checkout.
- Device support is environment-specific; no device works with every environment.
- Requires a configured .venv, i4h-workflows, Isaac Sim environment, and available teleoperation hardware.
- It cannot produce valid demonstrations without a human operator; it can only verify readiness in that situation.
- Device-specific controls appear only after launch, and the supplied skill does not provide complete movement tables for each device.
- A relative --record-to path can place the recording in an unintended nested directory.
How does this skill compare with similar options?
Side by side with related skills; every score comes from the same FSRS standard.
| Skill | FS score | Stars | Last updated | License |
|---|---|---|---|---|
| I4H Teleoperation Dataset Recorder this page ✓ NVIDIA · Official | 48 · Use with care | ★ 3.8k | today | Apache-2.0 |
| I4H Dataset Converter ✓ NVIDIA · Official | 40 · Not recommended | ★ 3.8k | today | Apache-2.0 |
| I4H Mimic Dataset Augmentation ✓ NVIDIA · Official | 52 · Use with care | ★ 3.8k | today | Apache-2.0 |
| i4H Dataset Annotation Workflow ✓ NVIDIA · Official | 44 · Not recommended | ★ 3.8k | today | Apache-2.0 |
| LeRobot Dataset Visualizer ✓ NVIDIA · Official | 49 · Use with care | ★ 3.8k | today | Apache-2.0 |
How did FollowSkills review this skill?
The skill requires human operation, prohibits headless teleop by default, calls for clean shutdown, and specifies output files and logs. However, it can clone a remote repository by default, create directories/symlinks, and launch GUI/hardware workflows without explicit pre-execution confirmation; sensitive-data handling, permission boundaries, and rollback are incomplete, so points are deducted.
The ordered procedure, device constraints, absolute-path requirement, log extraction, success criteria, and troubleshooting are fairly specific. However, it depends on an external repository, environment, hardware, and tmux session, with no executable tests or static reproducibility evidence for this skill's key paths; the score is conservatively limited under static calibration.
Trigger conditions, audience, environment examples, device mappings, non-fit cases, and no-operator behavior are documented. Chinese-language interaction support is not addressed, and the default GitHub clone lacks a mainland-China reachability assessment or alternative, so points are deducted.
The skill provides purpose, version 0.6.0, Apache-2.0 license, author, prerequisites, limitations, troubleshooting, and final-output requirements. It lacks the recommended Instructions and Examples sections, and changelog, maintenance ownership, and update-path details are incomplete; setup also depends on another skill, so points are deducted.
It defines a usable workflow for creating a run directory, launching teleop, saving HDF5, surfacing device-specific controls, and verifying successful episode counts. Actual completion depends on a human, GUI, robot environment, and hardware; the two-task benchmark is positive-only and reports 46% Codex effectiveness, so the static-calibrated score remains limited.
The files include a revisioned benchmark with evaluation tasks, metrics, and reported results, plus explicit file/log verification checks. They do not include raw logs, CI execution, or committed tests covering this skill's key paths, so the benchmark cannot be independently reproduced from the supplied files and points are deducted.
Open a dimension to read why it scored that way
Evidence confidence：Low — Mostly static review, author material or a limited demo; useful for discovery, not high-risk decisions.
See the full review method →