Overview

This project intends to run and build system to get a better reward from the CarRacingv0 environment game by using the Reinforcement Learning with Convolutional Reservoir Computing (RCRC) model. The RCRC model which combine the Convolutional Neural Networks (CNN) and Echo State Networks (ESN) model with a fixed random weight.

Students are required to demonstrate an ability to run the code and setup the game environment in addition to change the parameters by experience and optimize them to have a better performance and getting more rewards from the Car Racing game. This is the modified code from the original (https://github.com/Narsil/rl-baselines) and we added the Bayesian optimization to optimize the following parameters:

• Spectral radius in Echo state network
• The size of the Conv and ESN output weights
• The number of epochs

Running the code

The Car_Racing.ipynb file is a colab file to run and install all the requirements needed to run this code

Below is the youtube video showing how it is done and linking it to the google drive: https://youtu.be/76_Z9S5j82g

Note

We just modified and added the Bayesian Optimization for the rcrc.py code