mpSchrader/gym-sokoban
A Sokoban puzzle game environment for training and evaluating reinforcement learning agents.
Not currently ranked — collecting fresh signals.
star history
This repository provides an OpenAI Gym-compatible environment for the Sokoban puzzle game, designed specifically for reinforcement learning research. The environment generates random rooms to prevent neural networks from overfitting on predefined layouts. It implements the game rules from DeepMind’s Imagination Augmented Agents paper, serving as a testbed for training RL algorithms on spatial reasoning and planning tasks where irreversible mistakes can occur.
Frequently asked
- What is mpSchrader/gym-sokoban?
- A Sokoban puzzle game environment for training and evaluating reinforcement learning agents.
- Is gym-sokoban open source?
- Yes — mpSchrader/gym-sokoban is open source, released under the MIT license.
- What language is gym-sokoban written in?
- mpSchrader/gym-sokoban is primarily written in Python.
- How popular is gym-sokoban?
- mpSchrader/gym-sokoban has 406 stars on GitHub.
- Where can I find gym-sokoban?
- mpSchrader/gym-sokoban is on GitHub at https://github.com/mpSchrader/gym-sokoban.