Skip to main content

Public Benchmark surface

Environment catalog

Independently installable Benchmark distributions across control, planning, robotics, driving, and long-horizon games.

Research domains

Generated from the current catalog.

01

Navigation and planning

AtCoder · CodeChef · Gymnasium · Jumanji · MiniGrid

Collections
9
Task profiles
109
02

Control and robotics

DeepMind Control Suite · Gymnasium · Gymnasium-Robotics · MetaWorld · robosuite

Collections
7
Task profiles
139
03

Interactive simulation and games

ALE · ARC Prize · Crafter · HighwayEnv · Jackdaw · NLE · Stable-Retro · ViZDoom

Collections
8
Task profiles
54
01
AtCoder

AHC054 · Treant's Forest

Long-horizon constraint placement against a deterministic explorer with private targets.

02
AtCoder

AHC057 · Molecules

Schedule low-cost bonds among moving components on a toroidal plane.

04
CodeChef

WAREHOUS · Warehouseman

Full-range warehouse storage and retrieval through a compact forklift instruction language.

05
Gymnasium

Classic Control

Small state-space control tasks with discrete or scalar continuous Actions.

  • CartPole
  • Acrobot
  • Mountain Car
  • Continuous Mountain Car
  • Pendulum
06
Gymnasium

Toy Text

Rule-based discrete tasks for planning, stochastic transitions, and delayed reward.

  • Blackjack
  • CliffWalking
  • FrozenLake
  • Taxi
07
Gymnasium

Box2D

Landing, locomotion, and visual driving in medium-weight physics environments.

  • LunarLander
  • BipedalWalker
  • CarRacing
08
Gymnasium

MuJoCo

Continuous control from small arms and pendulums to contact-rich locomotion.

  • Reacher
  • Pusher
  • InvertedPendulum
  • InvertedDoublePendulum
  • Swimmer
  • Hopper
  • +5 more
09
DeepMind Control Suite

BENCHMARKING

Canonical state-based continuous-control tasks from balancing and locomotion to manipulation.

  • Acrobot / Cartpole / Pendulum
  • Ball in Cup / Finger
  • Cheetah / Hopper / Walker
  • Fish / Swimmer
  • Humanoid
  • Manipulator / Point Mass / Reacher
10
MiniGrid

MiniGrid

Partially observable navigation, memory, object interaction, and procedural layouts.

  • DoorKey
  • KeyCorridor
  • Memory
  • Fetch
  • MultiRoom
  • DynamicObstacles
  • +15 more
11
MiniGrid

WFC

Wave Function Collapse layouts spanning mazes, rooms, dungeons, obstacles, and caves.

  • Maze
  • Rooms
  • Dungeon
  • Obstacles
  • Skew
  • 22 presets total
12
MiniGrid

BabyAI

Language-conditioned instruction following from atomic navigation to composite missions.

  • GoTo
  • Open
  • PickupPut
  • Unlock
  • Composite
  • 40 tasks total
13
Jumanji

Single-Policy tasks

Structured combinatorial environments with exact action masks and task-specific progress diagnostics.

  • 2048 / Graph Coloring / Minesweeper
  • Rubik's Cube / Sudoku / Sliding Tile
  • BinPack / FlatPack / Knapsack
  • JobShop / Tetris
  • CVRP / TSP / Maze
  • Snake / PacMan
14
Gymnasium-Robotics

Robotics

Goal-conditioned reaching, manipulation, maze, dexterous hand, and kitchen tasks.

  • Fetch
  • Point / Ant Maze
  • Adroit Hand
  • Shadow Hand
  • FrankaKitchen
15
MetaWorld

MT

Single-task and multi-task robotic manipulation collections.

  • MT1
  • MT10
  • MT50
  • Custom collections
16
robosuite

Manipulation

Single- and dual-arm Panda manipulation with dense task state and per-Episode visual evidence.

  • Lift / Stack
  • NutAssembly
  • PickPlace
  • Door / Wipe / ToolHang
  • TwoArmLift / TwoArmPegInHole
  • TwoArmHandover / TwoArmTransport
17
HighwayEnv

Driving

Traffic negotiation, parking, racing, and lane keeping.

  • Highway
  • Merge
  • Roundabout
  • Intersection
  • TwoWay
  • Exit
  • +4 more
18
Crafter

Crafter

Open-world survival, resource collection, crafting, and long-horizon development from RGB observations.

  • Canonical achievement score
  • Long-horizon development
  • Survival development
19
ARC Prize

ARC-AGI-3

Interactive visual reasoning across complete games, in-Episode level resets, and official scorecard aggregation.

  • 25 version-pinned public games
  • Custom fixed collections
21
ALE

Atari

Arcade control using the redistributable Tetris ROM.

  • Tetris-v5
22
ViZDoom

Standard scenarios

First-person navigation and combat across the standard packaged scenarios.

  • Basic
  • DeadlyCorridor
  • Deathmatch
  • HealthGathering
  • MyWayHome
  • TakeCover
  • +1 more
23
Stable-Retro

Airstriker

Genesis arcade control from the packaged Level 1 state.

  • Airstriker-Genesis-v0
24
Jackdaw

Balatro

Long-horizon deckbuilding with hands, shops, Jokers, consumables, packs, and antes.

  • Red Deck · White Stake

Authoring boundary

Benchmarks live outside the Kernel.

External distributions use only the public authoring SPI and own their domain semantics, dependencies, Feedback, and conformance tests.

Read the authoring guide