The OpenAI Gymnasium is a reinforcement learning playground, with MDP problems ranging from simple to complicated single agent environments. The following report has been prepared by Cameron Braatz as an academic project for CSPB 3202: Intro to Artificial Intelligence at CU-Boulder.
For this project, I decided to go with Mountain Car from the Classic Control library. According to the documentation, the Mountain Car MDP is a deterministic problem that sees a cart placed stochastically at the bottom of a sinusoidal valley. The goal of which is to teach our cart-agent to climb the hill on the right, reaching the goal state (ie: flag).
The agent does not have the sheer power to drive directly uphill to the flag, instead needing to gather its momentum before making a successful ascent. See demonstration below.
Attempt7.mp4
Please refer to IntroToGymnasium.ipynb for the complete project notebook. Note: Video files do not seem to be rendering correctly in the GitHub environment, please see https://github.com/cambraatz/MountainCart/assets for all figures and videos used in the Python notebook.