Runs the Isaac for Healthcare end-to-end agentic pipeline from recording through validation.
npx skills add https://github.com/NVIDIA/skills --skill i4h-workflow-e2e
Please help me install this Agent Skill. Skill: i4h-workflow-e2e Source: https://github.com/NVIDIA/skills/tree/main/skills/i4h-workflow-e2e Command: npx skills add https://github.com/NVIDIA/skills --skill i4h-workflow-e2e
Running the full Isaac for Healthcare robot-learning pipeline requires coordinating nine stages (record, mimic, annotate, replay, convert, visualize, finetune, validate, summary) with correct environment configuration, resume logic, and monitoring, which is error-prone when done manually.
Ask your agent to run the full end-to-end smoke pipeline for the scissor pick-and-place task using the i4h workflow.
High-intent search queries and answers about installing and using this skill.
Run workflows/agentic/scripts/e2e/run.sh with --env <env> from the i4h-workflows repository root. The script executes all stages in order: setup, record, mimic, annotate, replay, convert, viz, finetune, validate, and summary.
Pass both --from-stage <stage> and --run-dir <existing-run> to run.sh to resume from a prior run. Using --from-stage without a matching --run-dir will fail because the script cannot locate the previous outputs.
Yes, run.sh supports --skip-mimic, --skip-annotate, --skip-replay, and --skip-viz flags. Note that assemble_trocar is inference-only, so finetune and checkpoint validation are skipped automatically for that environment.
This error means the workflow environment was not set up before running the pipeline. Run the i4h-workflow-setup skill first to create the .venv, since setup is the first pipeline stage and all later stages depend on it.
Track progress from a separate shell by tailing runs/.latest/logs/workflow.log for live per-stage output. Once the run finishes, read runs/.latest/logs/SUMMARY.txt for the final report.
The assemble_trocar environment is inference-only, so finetune and checkpoint validation are skipped and no checkpoint outputs are produced. Checkpoint artifacts exist only for trainable environments, and headless mode requires explicitly setting ARENA_HEADLESS=1.
tmuskalSet up the LongMemEval benchmarking environment with conda and HuggingFace datasets.
bensonlee5Execute SmolVLA vision-language-action policies for robotic manipulation tasks.
Wesley1600Automate build, test, and deployment workflows with modular Bash scripts.
ammonfifeControl Philips Hue lights, scenes, and rooms via a CLI interface.
