DemoCat

A tool based application for collecting human demonstration data from simulated and real-world environments.

---

Introduction

• 2024/09/19: Our project is currently under active development. Contributions via pull requests (PRs) are warmly welcome! 😊
• 2024/11/11: Added support for the Meta-World environment.
• 2025/10/01: Added multi-camera support.
• 2025/10/02: Added support for the LIBERO environment.

---

Installation

Different environments require specific installation procedures. Please follow the instructions below according to your target environment.

Meta-World

conda create -n democat_metaworld python=3.11 -y
conda activate democat_metaworld
pip install -r requirements.txt
pip install metaworld

LIBERO

conda create -n democat_libero python=3.8 -y
conda activate democat_libero
pip install -r requirements.txt
git clone https://github.com/Lifelong-Robot-Learning/LIBERO.git third_party/LIBERO
cd third_party/LIBERO
pip install -r requirements.txt
pip install -e .

Note for Windows users (LIBERO only):
You may need to install a patched version of egl_probe to ensure compatibility:

pip install https://github.com/mhandb/egl_probe/archive/fix_windows_build.zip

---

Running the Application

To launch the GUI for human demonstration collection:

python gui.py

This will open the main interface, which includes two primary modules:

Demonstration Collection Module

This module allows you to collect demonstrations using predefined keyboard controls.

Keyboard Shortcuts:

• p: Pause recording
• q: Start/resume recording
• e: Save the current trajectory

Demonstration Management Module

This module enables you to view, delete, and manage previously collected demonstrations.

---

Adding a New Environment

To integrate a new environment into DemoCat, follow these steps:

1. Create an environment folder under environments/, and implement a class that inherits from BaseEnv (defined in environments/base_env.py).

   • Example implementations:
     • environments/libero/libero_env.py
     • environments/metaworld/metaworld_env.py

2. Register the new environment and its tasks in the environment manager.
   Edit app/environment.py and update the _register_default_environments method in EnvironmentManager:

   # Example: Registering Meta-World tasks
   self.register_environment('metaworld_mt10')  # Environment name
   self.register_task('metaworld_mt10', 'Reach', Reach)
   self.register_task('metaworld_mt10', 'Push', Push)
   # ... add more tasks as needed
   
   # Example: Registering LIBERO tasks
   self.register_environment('libero_object')
   self.register_task('libero_object', 'Task1', libero_object_task_0)
   self.register_task('libero_object', 'Task2', libero_object_task_1)

Ensure your new environment follows the same interface and data format conventions to maintain compatibility with the rest of the system.