GenManip Documentation

GenManip is a robotics simulation benchmark developed using Isaac Sim. It currently supports GenManip-Bench for evaluating reasoning models and prompt-based frameworks in complex manipulation tasks.

This documentation provides guides for installation, usage, tutorials, API references, and information about the baseline methods provided.

Codebase Overview

Here is a high-level overview of the project’s directory structure:

• Directoryassets/

  • Directoryconfigs/

    • default.json # Core configuration for paths
  • Directoryrobots/

    • Directoryconfigs/

      • …
    • Directorypanda/

      • …
• Directorybaselines/

  • Directorymodular_framework/ # Baseline: Modular Manipulation System

    • Directoryconfigs/

      • …
    • Directorydeploy_sim/

      • …
    • Directorymodular_framework/

      • …
  • Directorylearning_based_framework/ # Baseline: Learning-based Manipulation

    • DirectoryGR1/

      • …
• Directoryconfigs/ # Simulation and task configurations

  • Directorycameras/

    • default.yml
  • Directoryeval/

    • DirectoryGenManipBench_Expanded/

      • …
  • Directoryrobots/

    • …
  • Directorytasks/

    • genmanipbench.yml
• Directorygenmanip/ # Core library code

  • …
• Directorygenmanip_bench/ # Benchmark specific code

  • …
• Directoryscripts/

  • eval_mf.sh # Example script for evaluating modular framework
• Directorystandalone_tools/ # Helper scripts

  • fake_port.py # Simple socket server for testing communication
  • usda_gen.py # Asset post-processing script
• .gitignore
• combine_configs.py # Script to merge configurations
• eval_V3.py # Main evaluation script
• render_test.py # Script to test rendering and installation

Navigate through the sidebar to find detailed information on specific topics.