MiniGrid DoorKey 5×5
A grid-world benchmark for learning a sequence of actions involving a key, a locked door, and a goal.
Published by Farama Foundation
Setup and evaluation
The agent has to find a key, open the door, and then reach the goal. It is a compact test of exploration and multi-step decisions under partial observability. Test on several seeds to check whether a policy learned the sequence across different layouts.
Install with pip install minigrid. Import gymnasium and minigrid, then create gymnasium.make('MiniGrid-DoorKey-5x5-v0'). Specify which observation wrappers you use when comparing results.
- Framework
- MiniGrid
- Access
- Local MiniGrid environment using Gymnasium. No account required.