We have made a tool to create and simulate simple Markov Decision Processes (MDP) and run Reinforcement Learning (RL) algorithms on them to help boost your intuition for those algorithms. Students often face difficulty in grasping the complex dynamics of RL algorithms and the effect hyperparameters have on agent learning. This tool is targeted toward students enrolled in introductory to advanced courses in Artificial Intelligence that contain concepts from reinforcement learning. The primary learning objective is to assist students in understanding the core concepts of RL, including value estimation, policy optimization, and exploration-exploitation trade-offs, in a hands-on, gamified manner. Through this tool, students can design custom 2D-grid MDPs, change the rewards, and modify hyperparameters such as learning rates, discount factors, and many more. They can visualize how their chosen environments and parameters impact the convergence rate and performance of the RL algorithms.
Project on GitHub: AbhishekVarghese/Markov-Decision-Problem-Simulation
Credits :
Music: Hanami Matsuri by Fabian Measures
Voice lines from the amazing text to voice by app.fliki.ai
Images borrowed from Slides by Dan Klein and Pieter Abbeel from UC Berkley
Prezi for their amazing presentation software!
Chat-GPT for coming up with the initial draft for this video description!