---
title: Lerobot Pusht Trainer
emoji: 🏆
colorFrom: indigo
colorTo: pink
sdk: gradio
sdk_version: 5.47.2
app_file: app.py
pinned: false
---


# LeRobot: Diffusion-Based Policy Trainer

This project demonstrates **Diffusion Policy** applied to robotics. It implements a training and inference pipeline for the **PushT** environment using Hugging Face's `lerobot` library.

The goal is to train a robot agent to push a T-shaped block to a target location using generative diffusion models, which offer superior stability and multimodal action distribution compared to traditional policies.

## Features

- **Diffusion Policy:** Uses Denoising Diffusion Probabilistic Models (DDPM) to generate robot action trajectories.
- **PushT Environment:** A 2D simulation benchmark for precise robotic manipulation.
- **Interactive UI:** A **Gradio** web interface to visualize the agent's performance in real-time.
- **PyTorch Backend:** Optimized training loop using standard PyTorch and Hugging Face accelerators.

## Tech Stack

- **Core:** Python, PyTorch
- **Library:** Hugging Face LeRobot
- **Interface:** Gradio
- **Environment:** Gym / PushT

## Installation

To run this locally, you need Python 3.10+ and the required dependencies.

```bash
# Clone the repository
git clone [https://huggingface.co/spaces/zino36/lerobot-pusht-trainer](https://huggingface.co/spaces/zino36/lerobot-pusht-trainer)
cd lerobot-pusht-trainer

# Install dependencies
pip install -r requirements.txt

Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference