Gymnasium Robotics Tutorial

This project is a hands-on exploration of robotics environments using the Gymnasium library, with a focus on MuJoCo-based environments. The goal is to learn how to set up, train, and evaluate reinforcement learning agents in these simulated robotic tasks.

Project Goal

• Understand the fundamentals of Gymnasium for robotics.
• Implement and train RL agents (e.g., using Stable Baselines3) on various robotics tasks.
• Learn how to evaluate the performance of trained agents.
• Explore different robotics environments like Fetch, ShadowHand, etc.

Technologies Used

• Python
• Gymnasium
• MuJoCo
• Stable Baselines3
• NumPy

Setup

Prerequisites

• Python 3.8+
• Git

Installation

1. Clone the repository (if you haven't already or for others to use):

   git clone https://github.com/shubhamt2897/Gymnasium_Robotics_Tutorial.git
   cd Gymnasium_Robotics_Tutorial

2. Create a virtual environment (recommended):

   python -m venv .venv
   source .venv/bin/activate  # On Windows use `.venv\Scripts\activate`

3. Install dependencies: A requirements.txt file will be added here. For now, you can install the core libraries:

   pip install gymnasium[robotics] stable-baselines3[extra]
   # Add other specific libraries like moviepy if needed for recording
   # pip install moviepy

   (Consider creating a requirements.txt file soon by running pip freeze > requirements.txt after installing your base packages).

Usage

This section will be updated as the project progresses. It will include instructions on:

• How to run training scripts (e.g., python train_agent.py --env <environment_id>).
• How to run evaluation scripts (e.g., python evaluate_agent.py --model_path <path_to_model> --env <environment_id>).
• How to use any Jupyter notebooks for experimentation.

Example (Placeholder)

# Placeholder for a quick example of how to load an environment
import gymnasium as gym

env = gym.make("FetchPickAndPlace-v3", render_mode="human") # Or any other env
observation, info = env.reset(seed=42)

for _ in range(1000):
    action = env.action_space.sample()  # agent policy is here
    observation, reward, terminated, truncated, info = env.step(action)

    if terminated or truncated:
        observation, info = env.reset()
env.close()

Current Status

• Project initialized.
• Basic dependencies identified.
• Focusing on setting up the environment and initial training/evaluation scripts.
• Currently exploring ["FetchPickAndPlace-v3", "FetchSlide"and MaMuJoCo env].

Next Steps / To-Do

• Develop a script for training an agent on a chosen robotics task.
• Develop a script for evaluating a trained agent.
• Add detailed instructions for running training and evaluation.
• Document results and observations.
• Explore more complex environments or tasks.

Contributing

Pull requests are welcome. For major changes, please open an issue first to discuss what you would like to change.