Rubix

29 Jul 2023

Project Link: Rubix on GitHub

Rubix is a deep reinforcement learning Rubik’s Cube solver written in Jax and Haiku.

Environment

The custom environment is developed in the style of the environments in Jumanji.

Disclaimer: Since starting this repository, InstaDeep has released Jumanji 0.2.0, which includes a RubiksCube environment.

Agents

This repository currently supports DQN, QR-DQN, and a discretized PPO agent. The implementations of the DQN-based agents are inspired by the DQN Zoo implementations.

Code Structure

The code is structured in a modular way to allow easy extensions for both the environment and agents. The primary focus has been to ensure that the Rubik’s Cube environment is both efficient and adaptable for future developments in reinforcement learning research.

---

For more detailed information, you can explore the GitHub repository.