Isaac Lab Extension Template for Mobile Manipulation: Door Opening Task

This repository provides a template for setting up a custom Isaac Lab extension for mobile manipulation, focusing on door-opening tasks. It integrates reinforcement learning (RL) with environments designed for robotics, as well as tools for simulation and control.

Prerequisites

Before using this template, ensure the following dependencies are installed:

• Isaac Lab
• Isaac Sim
• Python 3.10
• NVIDIA GPU (for simulation and RL training)
• USD files (Mobile Robot, Door with Lever): Download from Google Drive and place them in the appropriate directory.
  • You should modify usd_path for mobile robot, floating-base robot and door.

Installation

1. Make sure Isaac Sim and Isaac Lab is installed and properly configured.

2. Prepare a python interpreter (choose one option):

   Option 1: Use the existing Python bundled with Isaac Lab

   As explained in the documentation, the executable isaaclab.sh fetches the bundled python. To execute Python scripts, use:

   path-to-isaac-lab/isaaclab.sh -p [.py]

   Option 2: Set up a new conda environment

   path-to-isaac-lab/isaaclab.sh --conda [env_name]    # to create conda env
   
   conda activate [env_name]
   isaaclab -i                                         # to install isaac lab extensions in conda env

3. Clone this repository and install the extension library:

   git clone https://github.com/soom1017/isaaclab_door_open.git
   
   cd exts/soomin
   python -m pip install -e .      # if option1, path-to-isaac-lab/isaaclab.sh -p -m pip install -e .

Usage

Training

To train the robot to grasp the door using the PPO algorithm from RSL-RL:

python scripts/rsl_rl/train.py --task Template-Isaac-Open-Door-Franka-v0 --num_envs 64

Evaluation

To evaluate a pre-trained model:

python scripts/rsl_rl/play.py --task Template-Isaac-Open-Door-Franka-Play-v0 --num_envs 1

Experiment Results

• Tensorboard: Monitor training progress with Tensorboard.

  

• Play Video: Showcase the robot interacting with the environment.

  experiment_result.mp4