Mujoco Humanoid Reinforcement Learning Experiment

This project implements a reinforcement learning algorithm to teach a Mujoco Humanoid to stand up using Stable Baselines 3.

Converting to Jupyter Notebook for Colab

To convert the humanoid_rl.py file to a Jupyter notebook for Google Colab:

1. Upload the file to Google Colab:

   • Go to Google Colab
   • Click on File > Upload notebook > Choose file and select the humanoid_rl.py file

2. OR manually create a new notebook and copy each section between the # %% markers as separate cells:

   • Code blocks should be added as code cells
   • Comments/markdown sections (starting with # %% [markdown]) should be added as text cells with the # and [markdown] removed

3. Add this at the beginning of your notebook to install dependencies:

   !pip install gymnasium[mujoco]==0.28.1
   !pip install stable-baselines3==1.8.0

Experiment Overview

The experiment trains a PPO (Proximal Policy Optimization) agent on the Humanoid-Standup-v4 environment from Gymnasium.

Three parameters are varied to evaluate their impact on performance:

1. Learning rate
2. Network architecture (policy network size)
3. Entropy coefficient

Results are visualized with training curves and performance metrics.