Mountain Car Problem

GitHub Repository

In this project, is available a practical Demonstration of three different Reinforcement Learning Approaches which can be used in order to solve the Open AI Gym Mountain Car Problem. The Mountain Car problem is a classic Reinforcement Learning exercise. In this scenario, the agent (a car) is stuck in a valley and aims to drive up to the top of a hill by optimising it’s velocity and position (continuous state space). In this exercise, three different approaches have been considered in order to solve this problem: using tabular Q-learning, approximating the tabular representation using RBF and learning the weights of the RBF approximation on-line using Q-learning.

Tabular approximation, RBF Tabular approximation and Q-learning RBF approximation in action

In this section, is available a video demonstration of the 3 proposed methods in action in order to solve the mountain car problem. As shown in the following video, the Q-learning RBF approximation was the method which performed best in order to solve this task with the less amount of steps as possible.

RBF Q-learning Approximation optimised Cost-to-go function

Contacts

If you want to keep updated with my latest articles and projects follow me on Medium and subscribe to my mailing list. These are some of my contacts details:

Buy Me A Coffee

Subscribe to my mailing list to be updated about my new blog posts!

* indicates required