Skip to content

Repository: https://github.com/isaac-for-healthcare/i4h-workflows

End-to-end workflows from simulation to real-world deployment

Workflow Layout and Specialty Catalog#

This page describes the workflow layout and specialty catalog.

A Scene defines the simulated world. A Task defines reusable behavior. A Workflow selects one Scene and connects it to the Tasks needed to complete a goal.

Available Workflows#

Agent-ready Workflows#

The workflow repository supports agent-assisted setup, execution, authoring, and validation through natural-language prompts. These workflows can run with supported coding agents or with the repository's locally hosted agent stack.

For local inference, the Local i4h Agent runs NVIDIA Nemotron with SGLang and OpenCode, keeping workflow execution on your own NVIDIA GPU or DGX Spark. See Use with an AI Agent for agent-ready examples and instructions.

Layout#

Workflow definitions are grouped by clinical robotics specialty:

workflows/
├── i4h_workflows/
│   ├── laparoscopic-robotics/
│   ├── ultrasound-robotics/
│   ├── endoluminal-robotics/
│   └── hospital-automation-robotics/
└── i4h_workflow_modes/
Specialty Workflows
Laparoscopic robotics surgical_lift_block, surgical_lift_needle, surgical_lift_needle_organs, surgical_reach_dual_psm, surgical_reach_psm, surgical_reach_star
Ultrasound robotics ultrasound_liver_scan, ultrasound_probe_reach
Endoluminal robotics endoluminal_navigation
Hospital automation robotics assemble_trocar, locomanip_push_cart, locomanip_tray_pick_and_place, scissor_pick_and_place

The Python filename is the public workflow ID and must be unique across all specialties. Specialty folders organize the source only. Runtime commands use the workflow ID without the specialty path:

./run.sh surgical_reach_psm --rule-based

Standard run-mode builders shared by multiple Workflows live in i4h_workflow_modes/.