Minigrid

We use minigrid==3.0.0 as the codebase. See https://github.com/Farama-Foundation/Minigrid/tree/v3.0.0

EnvPool supports all 81 non-BabyAI MiniGrid-* environments and all 96 BabyAI-* environments registered by upstream minigrid==3.0.0.

Render Compare

Representative first-frame compares for the supported MiniGrid and BabyAI tasks. In each panel, EnvPool is on the left and upstream minigrid is on the right.

../_images/minigrid_official_compare.png ../_images/babyai_official_compare.png

Options

  • task_id (str): see the available tasks below;

  • num_envs (int): how many environments you would like to create;

  • batch_size (int): the expected batch size for return result, default to num_envs;

  • num_threads (int): the maximum thread number for executing the actual env.step, default to batch_size;

  • seed (int | Sequence[int]): the environment seed. When a sequence is provided, it must contain exactly one seed per environment. Default to 42;

  • max_episode_steps (int): the maximum number of steps for one episode. The default value depends on task_id and follows the upstream MiniGrid registration.

Observation Space

Each MiniGrid observation contains:

  • obs["image"]: a (agent_view_size, agent_view_size, 3) uint8 tensor using the standard MiniGrid object/color/state encoding;

  • obs["direction"]: the agent direction in [0, 3];

  • obs["mission"]: a fixed-size uint8 byte buffer with length 96 for MiniGrid-* tasks and 512 for BabyAI-* tasks;

  • info["agent_pos"]: the agent position in the full grid;

  • info["mission_id"]: a stable integer ID when the mission comes from a finite canonical set, otherwise -1.

Use envpool.minigrid.decode_mission(...) to decode the mission buffer back to a Python string:

import envpool
from envpool.minigrid import decode_mission

env = envpool.make_gymnasium("MiniGrid-DoorKey-8x8-v0", num_envs=1)
obs, info = env.reset()
mission = decode_mission(obs["mission"][0])

Action Space

Most tasks expose the standard MiniGrid discrete action space with values in [0, 6]:

  • 0: turn left

  • 1: turn right

  • 2: move forward

  • 3: pick up an object

  • 4: drop an object

  • 5: toggle / interact

  • 6: done

MiniGrid-Dynamic-Obstacles-* follows upstream and only uses the movement subset [0, 2].

Available Tasks

All upstream BabyAI-* task IDs from minigrid==3.0.0 are available in addition to the MiniGrid-* task IDs listed below.

Empty

  • MiniGrid-Empty-5x5-v0

  • MiniGrid-Empty-Random-5x5-v0

  • MiniGrid-Empty-6x6-v0

  • MiniGrid-Empty-Random-6x6-v0

  • MiniGrid-Empty-8x8-v0

  • MiniGrid-Empty-16x16-v0

DoorKey

  • MiniGrid-DoorKey-5x5-v0

  • MiniGrid-DoorKey-6x6-v0

  • MiniGrid-DoorKey-8x8-v0

  • MiniGrid-DoorKey-16x16-v0

DistShift

  • MiniGrid-DistShift1-v0

  • MiniGrid-DistShift2-v0

Crossing

  • MiniGrid-LavaCrossingS9N1-v0

  • MiniGrid-LavaCrossingS9N2-v0

  • MiniGrid-LavaCrossingS9N3-v0

  • MiniGrid-LavaCrossingS11N5-v0

  • MiniGrid-SimpleCrossingS9N1-v0

  • MiniGrid-SimpleCrossingS9N2-v0

  • MiniGrid-SimpleCrossingS9N3-v0

  • MiniGrid-SimpleCrossingS11N5-v0

LavaGap

  • MiniGrid-LavaGapS5-v0

  • MiniGrid-LavaGapS6-v0

  • MiniGrid-LavaGapS7-v0

Dynamic Obstacles

  • MiniGrid-Dynamic-Obstacles-5x5-v0

  • MiniGrid-Dynamic-Obstacles-Random-5x5-v0

  • MiniGrid-Dynamic-Obstacles-6x6-v0

  • MiniGrid-Dynamic-Obstacles-Random-6x6-v0

  • MiniGrid-Dynamic-Obstacles-8x8-v0

  • MiniGrid-Dynamic-Obstacles-16x16-v0

Fetch

  • MiniGrid-Fetch-5x5-N2-v0

  • MiniGrid-Fetch-6x6-N2-v0

  • MiniGrid-Fetch-8x8-N3-v0

FourRooms

  • MiniGrid-FourRooms-v0

GoToDoor

  • MiniGrid-GoToDoor-5x5-v0

  • MiniGrid-GoToDoor-6x6-v0

  • MiniGrid-GoToDoor-8x8-v0

GoToObject

  • MiniGrid-GoToObject-6x6-N2-v0

  • MiniGrid-GoToObject-8x8-N2-v0

KeyCorridor

  • MiniGrid-KeyCorridorS3R1-v0

  • MiniGrid-KeyCorridorS3R2-v0

  • MiniGrid-KeyCorridorS3R3-v0

  • MiniGrid-KeyCorridorS4R3-v0

  • MiniGrid-KeyCorridorS5R3-v0

  • MiniGrid-KeyCorridorS6R3-v0

LockedRoom

  • MiniGrid-LockedRoom-v0

Memory

  • MiniGrid-MemoryS17Random-v0

  • MiniGrid-MemoryS13Random-v0

  • MiniGrid-MemoryS13-v0

  • MiniGrid-MemoryS11-v0

  • MiniGrid-MemoryS9-v0

  • MiniGrid-MemoryS7-v0

MultiRoom

  • MiniGrid-MultiRoom-N2-S4-v0

  • MiniGrid-MultiRoom-N4-S5-v0

  • MiniGrid-MultiRoom-N6-v0

ObstructedMaze

  • MiniGrid-ObstructedMaze-1Dl-v0

  • MiniGrid-ObstructedMaze-1Dlh-v0

  • MiniGrid-ObstructedMaze-1Dlhb-v0

  • MiniGrid-ObstructedMaze-2Dl-v0

  • MiniGrid-ObstructedMaze-2Dlh-v0

  • MiniGrid-ObstructedMaze-2Dlhb-v0

  • MiniGrid-ObstructedMaze-1Q-v0

  • MiniGrid-ObstructedMaze-2Q-v0

  • MiniGrid-ObstructedMaze-Full-v0

  • MiniGrid-ObstructedMaze-2Dlhb-v1

  • MiniGrid-ObstructedMaze-1Q-v1

  • MiniGrid-ObstructedMaze-2Q-v1

  • MiniGrid-ObstructedMaze-Full-v1

Playground

  • MiniGrid-Playground-v0

PutNear

  • MiniGrid-PutNear-6x6-N2-v0

  • MiniGrid-PutNear-8x8-N3-v0

RedBlueDoors

  • MiniGrid-RedBlueDoors-6x6-v0

  • MiniGrid-RedBlueDoors-8x8-v0

Unlock

  • MiniGrid-Unlock-v0

UnlockPickup

  • MiniGrid-UnlockPickup-v0

BlockedUnlockPickup

  • MiniGrid-BlockedUnlockPickup-v0

WFC

Representative first-frame compares for the WFC presets below. In each row, EnvPool is on the left and upstream minigrid is on the right.

../_images/minigrid_wfc_render.png

  • MiniGrid-WFC-MazeSimple-v0

  • MiniGrid-WFC-DungeonMazeScaled-v0

  • MiniGrid-WFC-RoomsFabric-v0

  • MiniGrid-WFC-ObstaclesBlackdots-v0

  • MiniGrid-WFC-ObstaclesAngular-v0

  • MiniGrid-WFC-ObstaclesHogs3-v0

Validation

All registered MiniGrid task IDs are covered by:

  • //envpool/minigrid:minigrid_align_test for upstream behavioral alignment. Dynamic Obstacles is aligned by transition replay rather than by sharing the exact same RNG bitstream, because upstream NumPy uses PCG64 while EnvPool uses C++ mt19937;

  • //envpool/minigrid:minigrid_deterministic_test for same-seed determinism;

  • //envpool:make_test for top-level construction coverage through the public envpool.make_* entry points.