Skip to content

Robotic Surgery

Complete surgical automation framework combining simulation, AI training, and deployment for robotic surgical systems.

Overview

The robotic surgery workflow provides a comprehensive framework for developing and deploying autonomous surgical capabilities. Built on Isaac Sim and Isaac Lab, it enables researchers and developers to:

  • Train AI policies for surgical manipulation tasks
  • Simulate complex tissue interactions with realistic physics
  • Deploy trained models to real surgical robots
  • Evaluate performance with clinical metrics

What's Included

High-fidelity surgical scene rendering
Deformable tissue simulation powered by NVIDIA PhysX with realistic material properties

Pre-built environments
State machine implementations and RL environments for fundamental surgical tasks

Development tools
Performance profiling, trajectory visualization, and debugging utilities

Hardware support
Compatible with dVRK (da Vinci Research Kit), Virtual Incision MIRA, Universal Robots arms, and custom robot integration

Get Started

  • Quick Start Guide

    Set up your development environment and run your first surgical simulation

    What you'll learn:

    • Install dependencies and drivers
    • Download required assets
    • Run example demonstrations
    • Understand the basic framework

    View Setup Instructions →

  • State Machine Environments

    Hand-crafted state machines for fundamental surgical subtasks

    Available tasks:

    • dVRK-PSM and STAR robot reaching
    • Dual-arm bimanual coordination
    • Suture needle lifting and manipulation
    • Peg block transfer tasks

    Explore State Machines →

  • Reinforcement Learning

    Train adaptive AI policies for surgical automation

    Training capabilities:

    • RSL-RL framework with PPO
    • Multi-GPU training support
    • dVRK-PSM reaching tasks
    • Suture needle manipulation

    Start Training →

Extend with Your Own Assets

The robotic surgery workflow integrates seamlessly with custom hardware and patient models: