Reinforcement Learning
Interactive Playground for Basic Reinforcement Learning Algorithms
Project Overview
This project implements implements a variety of basic reinforcement learning algorithms with adjustable hyperparameters to allow for easy experimentation and understanding of the algorithms. These algorithms are tasked to find an optimal path in a grid world maze-like environment. Play around with the algorithms and see how they perform throughout the training process with different hyperparameter settings and environment configurations!
Key Features
- Implementation of SARSA, Q-Learning, Actor Critic, SARSA-λ algorithms
- Customizable environments and hyperparameters to allow for easy experimentation
- Visualization of the training process and agent's performance
Technical Details
The reinforcement learning system is built on a modular architecture that allows for easy experimentation with different algorithms and environments. The core components include:
Custom Environments
Customizable grid world environment with various obstacles and to create complex mazes and environments for the models to navigate through.
Adjustable Models
A variety of RL models with adjustable hyperparameters to allow for easy experimentation with different algorithms in the same environment.
Training Visualization
Ability to scroll through the episodes of the training process and see the agent's performance at each step. Also plotting of the average reward per episode throughout the training process.
Try It Yourself
Build your own environment or generate a random one, then train a reinforcement learning agent to navigate through it.
Algorithm Settings
Environment Builder
Training Progress
In Development
Check back in the future for a Deep RL playground with more complex environments and more advanced algorithms!