LeNewton

LeNewton is a simulation environment for the SO100 robot arm, built on top of the Newton physics engine and NVIDIA Warp. It provides a high-performance, differentiable simulation platform for robotic manipulation tasks.

Installation

Prerequisites

• Python 3.11
• CUDA-capable GPU (recommended for Warp/Newton)

Install from Source

1. Clone the repository:

   git clone https://github.com/makolon/lenewton.git
   cd lenewton

2. Install the package:

   pip install -e .

Usage

Running Examples

You can find example scripts in the examples/ directory. To run the basic environment creation and block stacking task example:

python examples/01_create_env.py

This script demonstrates:

1. Creating a BlockStackTask.
2. Initializing the LeNewtonEnv with the SO100 robot.
3. Setting up the viewer and video recording.

Creating a Custom Task

Tasks are defined by inheriting from LeNewtonTask. See src/lenewton/tasks/block_stacking.py for a reference implementation.

from lenewton.env import LeNewtonEnv
from lenewton.tasks import BlockStackTask

# Define the task
task = BlockStackTask(
    task_name="BlockStack",
    goal_str="Stack the blocks",
    max_steps=500
)

# Initialize the environment
env = LeNewtonEnv(
    task=task,
    render_mode="rgb_array",
    use_viewer=True
)

# Simulation loop
obs = env.reset()
# ... action loop ...

Project Structure

lenewton/
├── examples/           # Example scripts
├── src/
│   └── lenewton/
│       ├── assets/     # Robot URDFs, meshes, and scene XMLs
│       ├── tasks/      # Task definitions (e.g., BlockStackTask)
│       ├── teleop/     # Teleoperation & data collection
│       ├── utils/      # Utilities for simulation
│       └── env.py      # Main environment class
├── pyproject.toml      # Project configuration and dependencies
└── README.md

License

This project is licensed under the terms of the included LICENSE file.