| Definition | SimpleGrid is customizable grid game environment for Gymnasium, designed for quickly testing and prototyping Reinforcement Learning algorithms. |
| No of environments | 102 |
SimpleGrid allows to navigate from a designated start tile to a goal tile while avoiding obstacles represented as walls. The environment supports both default grid sizes (4×4 and 8×8) and custom, random-generated maps, making it highly adaptable for different testing scenarios.
The agent operates within a discrete action space, where it can move up, down, left, or right on the grid. The observation space represents the agent’s position as a single integer, which can be converted to 2D coordinates for easier understanding. SimpleGrid also features a straightforward reward system that provides feedback based on the agent’s actions: positive rewards for reaching the goal, negative rewards for colliding with walls or attempting to exit the grid, and zero rewards for moving into empty spaces. This flexibility and simplicity make SimpleGrid an excellent choice for both educational purposes and practical experimentation in Reinforcement Learning.
| Example observations | |
| Example actions | |
| Example reward |