Skip to content

⚑⚑ Lightning Fast (~300TPS) Reinforcement Learning environment on latest Minecraft 🏝️

License

Notifications You must be signed in to change notification settings

yhs0602/CraftGround

Folders and files

NameName
Last commit message
Last commit date

Latest commit

Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 

Repository files navigation

CraftGround - Reinforcement Learning Environment for Minecraft

Wheels Python package CMake Build Gradle Build

CraftGround_Logo

Hits

RL experiments using lightweight minecraft environment

This is the latest development repository of CraftGround environment.

Quick start

  1. You need to install JDK >= 21
  2. Run the following command to install the package.
    pip install craftground
  3. Take a look at the the demo repository!
  4. Here is a simple example that uses this environment.
    from craftground import craftground
    from craftground.wrappers.action import ActionWrapper, Action
    from craftground.wrappers.fast_reset import FastResetWrapper
    from craftground.wrappers.time_limit import TimeLimitWrapper
    from craftground.wrappers.vision import VisionWrapper
    from stable_baselines3 import A2C
    from stable_baselines3.common.monitor import Monitor
    from stable_baselines3.common.vec_env import VecVideoRecorder, DummyVecEnv
    from wandb.integration.sb3 import WandbCallback
    
    import wandb
    from avoid_damage import AvoidDamageWrapper
    
    
    def main():
       run = wandb.init(
           # set the wandb project where this run will be logged
           project="craftground-sb3",
           entity="jourhyang123",
           # track hyperparameters and run metadata
           group="escape-husk",
           sync_tensorboard=True,  # auto-upload sb3's tensorboard metrics
           monitor_gym=True,  # auto-upload the videos of agents playing the game
           save_code=True,  # optional    save_code=True,  # optional
       )
       env = craftground.make(
           # env_path="../minecraft_env",
           port=8023,
           initialInventoryCommands=[],
           initialPosition=None,  # nullable
           initialMobsCommands=[
               "minecraft:husk ~ ~ ~5 {HandItems:[{Count:1,id:iron_shovel},{}]}",
               # player looks at south (positive Z) when spawn
           ],
           imageSizeX=114,
           imageSizeY=64,
           visibleSizeX=114,
           visibleSizeY=64,
           seed=12345,  # nullable
           allowMobSpawn=False,
           alwaysDay=True,
           alwaysNight=False,
           initialWeather="clear",  # nullable
           isHardCore=False,
           isWorldFlat=True,  # superflat world
           obs_keys=["sound_subtitles"],
           initialExtraCommands=[],
           isHudHidden=False,
           render_action=True,
           render_distance=2,
           simulation_distance=5,
       )
       env = FastResetWrapper(
           TimeLimitWrapper(
               ActionWrapper(
                   AvoidDamageWrapper(VisionWrapper(env, x_dim=114, y_dim=64)),
                   enabled_actions=[Action.FORWARD, Action.BACKWARD],
               ),
               max_timesteps=400,
           )
       )
       env = Monitor(env)
       env = DummyVecEnv([lambda: env])
       env = VecVideoRecorder(
           env,
           f"videos/{run.id}",
           record_video_trigger=lambda x: x % 2000 == 0,
           video_length=200,
       )
       model = A2C(
           "MlpPolicy", env, verbose=1, device="mps", tensorboard_log=f"runs/{run.id}"
       )
    
       model.learn(
           total_timesteps=10000,
           callback=WandbCallback(
               gradient_save_freq=100,
               model_save_path=f"models/{run.id}",
               verbose=2,
           ),
       )
       model.save("a2c_craftground")
       run.finish()
    
       # vec_env = model.get_env()
       # obs = vec_env.reset()
       # for i in range(1000):
       #     action, _state = model.predict(obs, deterministic=True)
       #     obs, reward, done, info = vec_env.step(action)
       #     # vec_env.render("human")
       #     # VecEnv resets automatically
       #     # if done:
       #     #   obs = vec_env.reset()
    
    
    if __name__ == "__main__":
       main()

Environment

https://github.com/yhs0602/MinecraftEnv

Utilizing protocol buffers, we've constructed a reinforcement learning environment specifically tailored for Minecraft. Below are detailed specifications of the environment's architecture:

Initial Environment

BlockState

Field Type Description
x, y, z int32 Coordinates of the block.
block_state string State of the block, e.g., minecraft:andesite_stairs[facing=east,half=bottom,shape=straight,waterlogged=false].

InitialEnvironmentMessage

Field Type Description
initialInventoryCommands repeated string Commands to setup initial inventory.
initialPosition repeated int32 Player's starting coordinates.
initialMobsCommands repeated string Commands for spawning mobs.
imageSizeX, imageSizeY int32 Image dimensions of the environment.
seed int64 World generation seed.
allowMobSpawn bool Controls mob spawning.
alwaysNight, alwaysDay bool Control for time of day.
initialWeather string Initial weather setting.
isWorldFlat bool Flat world toggle.
visibleSizeX, visibleSizeY int32 Player's visible dimensions.
initialExtraCommands repeated string Extra commands for initialization.
killedStatKeys, minedStatKeys, miscStatKeys repeated string Player's statistic keys.
initialBlockStates repeated BlockState Initial world block states.
surroundingEntityDistances repeated int32 Entity distances from player.
hudHidden bool Toggle for HUD visibility.
render_distance, simulation_distance int32 Block and entity render/simulation distances.

Observation Space

ItemStack

Field Type Description
raw_id int32 Unique item identifier.
translation_key string Item's display name translation key.
count int32 Amount in the item stack.
durability, max_durability int32 Durability information of item.

BlockInfo

Field Type Description
x, y, z int32 Block coordinates.
translation_key string Block's display name translation key.

EntityInfo

Field Type Description
unique_name string Unique name of the entity.
translation_key string Entity's translation key.
x, y, z double Entity coordinates.
yaw, pitch double Yaw and Pitch of the entity.
health double Health of the entity.

ObservationSpaceMessage

Field Type Description
image bytes Image data of the environment.
x, y, z double Player's coordinates in the world.
yaw, pitch double Player's orientation (yaw & pitch).
health double Player's health level.
food_level double Player's food level.
saturation_level double Player's saturation level.
is_dead bool Flag indicating if the player is dead.
inventory repeated ItemStack List of items in player's inventory.
raycast_result HitResult Raycasting result to identify targeted blocks or entities.
sound_subtitles repeated SoundEntry List of recent sounds with subtitles.
status_effects repeated StatusEffect List of active status effects on the player.
killed_statistics map<string, int32> Player's kill statistics with entity names as keys.
mined_statistics map<string, int32> Player's block mining statistics with block types as keys.
misc_statistics map<string, int32> Miscellaneous statistics.
visible_entities repeated EntityInfo List of entities currently visible to the player.
surrounding_entities map<int32, EntitiesWithinDistance> Map of entities around the player with distances as keys.
bobber_thrown bool Flag indicating if a fishing bobber is currently thrown.
world_time int64 Current world time.
last_death_message string Last death reason.

Action Space

ActionSpaceMessage

Field Type Description
action repeated int32 Available player actions.
commands repeated string Any minecraft commands.

Wrapper list

Wrapper Name Description
action Defines discrete action spaces and operations for the agent.
sound Provides sound-based feedback or actions for the agent.
vision Incorporates visual feedback or vision-based actions for the agent.

How to execute minecraft command in a gymnasium wrapper?

self.get_wrapper_attr("add_commands")(
    [
        f"setblock 1 2 3 minecraft:cake"
    ]
)

Devaju font license

https://dejavu-fonts.github.io/License.html

Adding a new paramerter

  1. Edit .proto
  2. protoc
  3. Edit python files

Formatting

Install formatters

brew install ktlint clang-format google-java-format
find . \( -iname '*.h' -o -iname '*.cpp' -o -iname '*.mm' \) | xargs clang-format -i
ktlint '!src/craftground/MinecraftEnv/src/main/java/com/kyhsgeekcode/minecraftenv/proto/**'
find . -name '*.java' -print0 | xargs -0 -P 4 google-java-format -i

Managing proto files

cd src/
protoc proto/action_space.proto --python_out=craftground
protoc proto/initial_environment.proto --python_out=craftground
protoc proto/observation_space.proto --python_out=craftground
protoc proto/action_space.proto --java_out=craftground/MinecraftEnv/src/main/java/ --kotlin_out=craftground/MinecraftEnv/src/main/java/
protoc proto/initial_environment.proto --java_out=craftground/MinecraftEnv/src/main/java/ --kotlin_out=craftground/MinecraftEnv/src/main/java/
protoc proto/observation_space.proto --java_out=craftground/MinecraftEnv/src/main/java/ --kotlin_out=craftground/MinecraftEnv/src/main/java/

Troubleshooting

Protobuf runtime version error

google.protobuf.runtime_version.VersionError: Detected incompatible Protobuf Gencode/Runtime versions when loading proto/initial_environment.proto: gencode 5.29.1 runtime 5.27.3. Runtime version cannot be older than the linked gencode version. See Protobuf version guarantees at https://protobuf.dev/support/cross-version-runtime-guarantee.

Solution

pip install --upgrade protobuf