Reinforcement Learning Project: MuJoCo Hopper and Panda-Gym PushTask

This repository contains a reinforcement learning project developed for the course Fundamentals of Artificial Intelligence, Machine Learning and Deep Learning at Politecnico di Torino.

The project focuses on training, evaluating, and analyzing reinforcement learning agents in simulated control environments. The main environments considered are MuJoCo Hopper and Panda-Gym PushTask.

The work includes environment setup, random policy testing, agent training, evaluation, reward analysis, and technical reporting.

Project Overview

The goal of the project is to study reinforcement learning methods for control tasks in simulation-based environments.

The repository is divided into two main parts:

• Part 1 — MuJoCo Hopper
  Experiments on the Hopper continuous-control environment using Gymnasium and MuJoCo.

• Part 2 — Panda-Gym PushTask
  Experiments on a robotic pushing task using Panda-Gym.

Main Topics

• Reinforcement Learning
• Continuous Control
• Policy Optimization
• Simulation-Based Learning
• Agent Evaluation
• Reward Analysis
• Domain Randomization
• Robotics Simulation

Tools and Libraries

• Python
• Gymnasium
• MuJoCo
• Stable-Baselines3
• Panda-Gym
• NumPy
• Matplotlib
• Jupyter Notebook

Repository Structure

. ├── part1/ # MuJoCo Hopper experiments ├── part2/ # Panda-Gym PushTask experiments ├── Project_RL.pdf # Technical project report ├── requirements.txt # Python dependencies └── README.md

Report

The full technical report is available here:

Reinforcement Learning Project Report⁠

The report describes the methodology, implementation choices, experiments, results, and limitations of the project.

How to Run

Install the required dependencies: pip install -r requirements.txt

Run the initial random-policy test for Part 1: cd part1 python test_random_policy.py

For the Panda-Gym task: cd part2/panda-gym pip install -e .

Then run the relevant training or evaluation scripts from the project folders.

Skills Demonstrated

• Reinforcement learning experimentation
• Python-based machine learning workflows
• Simulation environment setup
• Agent training and evaluation
• Reward curve analysis
• Technical reporting
• Scientific computing