def specs(self) -> Dict:
return {
"forward": spaces.Discrete(2),
"jump": spaces.Discrete(2),
"camera_pitch": spaces.Discrete(self._num_camera_actions),
"camera_yaw": spaces.Discrete(self._num_camera_actions),
"place": obtain_action_space["place"],
"equip": obtain_action_space["equip"],
"craft": obtain_action_space["craft"],
"nearbyCraft": obtain_action_space["nearbyCraft"],
"nearbySmelt": obtain_action_space["nearbySmelt"]
}
register(
id='MineRLTreechop-v0',
entry_point='minerl.env:MineRLEnv',
kwargs={
'xml':
os.path.join(missions_dir, 'treechop.xml'),
'observation_space':
spaces.Dict({
'pov':
spaces.Box(low=0, high=255, shape=(64, 64, 3), dtype=np.uint8),
}),
'action_space':
spaces.Dict({
"forward":
spaces.Discrete(2),
"back":
spaces.Discrete(2),
"left":
spaces.Discrete(2),
"right":
spaces.Discrete(2),
"jump":
spaces.Discrete(2),
"sneak":
spaces.Discrete(2),
"sprint":
spaces.Discrete(2),
"attack":
spaces.Discrete(2),
"camera":
def register(environment_id, xml_path):
"""
Given an xml_path and a desired environment_id,
registers the Minecraft mission as a gym environment.
The environment_id is a string which you "register" here
and then call later when doing gym.make(environment_id).
The xml_path is the path to the Malmo mission file, which is
in xml file format, of course.
"""
ENVIRONMENT_ID = environment_id
XML_PATH = xml_path
def make_navigate_text(top, dense):
"""
Helper function to construct the description for this gym environment.
TODO: Copied as an example and needs to be removed / changed
"""
navigate_text = """
.. image:: ../assets/navigate{}1.mp4.gif
:scale: 100 %
:alt:
.. image:: ../assets/navigate{}2.mp4.gif
:scale: 100 %
:alt:
.. image:: ../assets/navigate{}3.mp4.gif
:scale: 100 %
:alt:
.. image:: ../assets/navigate{}4.mp4.gif
:scale: 100 %
:alt:
In this task, the agent must move to a goal location denoted by a diamond block. This represents a basic primitive used in many tasks throughout Minecraft. In addition to standard observations, the agent has access to a “compass” observation, which points near the goal location, 64 meters from the start location. The goal has a small random horizontal offset from the compass location and may be slightly below surface level. On the goal location is a unique block, so the agent must find the final goal by searching based on local visual features.
The agent is given a sparse reward (+100 upon reaching the goal, at which point the episode terminates).
"""
if dense:
navigate_text += "**This variant of the environment is dense reward-shaped where the agent is given a reward every tick for how much closer (or negative reward for farther) the agent gets to the target.**\n"
else:
navigate_text += "**This variant of the environment is sparse.**\n"
if top is "normal":
navigate_text += "\nIn this environment, the agent spawns on a random survival map.\n"
navigate_text = navigate_text.format(*["" for _ in range(4)])
else:
navigate_text += "\nIn this environment, the agent spawns in an extreme hills biome.\n"
navigate_text = navigate_text.format(
*["extreme" for _ in range(4)])
return navigate_text
# Define the action space
navigate_action_space = spaces.Dict({
"forward":
spaces.Discrete(2),
"back":
spaces.Discrete(2),
"left":
spaces.Discrete(2),
"right":
spaces.Discrete(2),
"jump":
spaces.Discrete(2),
# "sneak": spaces.Discrete(2),
# "sprint": spaces.Discrete(2),
# "attack": spaces.Discrete(2),
"camera":
spaces.Box(low=-180, high=180, shape=(2, ), dtype=np.float32),
"place":
spaces.Enum('none', 'dirt')
})
# Define the information that the agent receives, and navigates using
navigate_observation_space = spaces.Dict({
'pov':
spaces.Box(low=0, high=255, shape=(64, 64, 3), dtype=np.uint8),
'inventory':
spaces.Dict(
spaces={
'dirt': spaces.Box(low=0, high=2304, shape=(), dtype=np.int)
}),
'compassAngle':
spaces.Box(low=-180.0, high=180.0, shape=(), dtype=np.float32)
})
# This is the function call where we actually register
# the environment
gym.envs.registration.register(
id=ENVIRONMENT_ID,
entry_point='minerl.env:MineRLEnv',
kwargs={
'xml': XML_PATH,
'observation_space': navigate_observation_space,
'action_space': navigate_action_space,
'docstr': make_navigate_text('normal', False)
},
max_episode_steps=6000,
)
import numpy as np
missions_dir = os.path.join(os.path.dirname(__file__), 'missions')
register(
id='MineRLWoolGatheringTrain-v0',
entry_point='minerl.env:MineRLEnv',
kwargs={
'xml': os.path.join(missions_dir, 'wool_gather_train.xml'),
'observation_space': spaces.Dict({
'pov': spaces.Box(low=0, high=255, shape=(64, 64, 3), dtype=np.uint8),
}),
'action_space': spaces.Dict(spaces={
"forward": spaces.Discrete(2),
"back": spaces.Discrete(2),
"left": spaces.Discrete(2),
"right": spaces.Discrete(2),
"jump": spaces.Discrete(2),
"sneak": spaces.Discrete(2),
"sprint": spaces.Discrete(2),
"attack": spaces.Discrete(2),
"camera": spaces.Box(low=-180, high=180, shape=(2,), dtype=np.float32),
}),
},
max_episode_steps=8000,
reward_threshold=1000.0,
)
register(
def __init__(self):
self.state = {
'equipped_items': {
'mainhand': {
'damage': 0,
'maxDamage': 0,
'type': 0
}
},
'inventory': {
'coal': 0,
'cobblestone': 0,
'crafting_table': 0,
'dirt': 0,
'furnace': 0,
'iron_axe': 0,
'iron_ingot': 0,
'iron_ore': 0,
'iron_pickaxe': 0,
'log': 0,
'planks': 0,
'stick': 0,
'stone': 0,
'stone_axe': 0,
'stone_pickaxe': 0,
'torch': 0,
'wooden_axe': 0,
'wooden_pickaxe': 0
},
'pov': np.full([64, 64, 3], 127, dtype=np.uint8)
}
self.action_space = {
'attack':
spaces.Discrete(2),
'back':
spaces.Discrete(2),
'camera':
spaces.Box(np.full([2], -1), np.full([2], 1)),
'craft':
spaces.Enum('none', 'torch', 'stick', 'planks', 'crafting_table'),
'equip':
spaces.Enum('none', 'air', 'wooden_axe', 'wooden_pickaxe',
'stone_axe', 'stone_pickaxe', 'iron_axe',
'iron_pickaxe'),
'forward':
spaces.Discrete(2),
'jump':
spaces.Discrete(2),
'left':
spaces.Discrete(2),
'nearbyCraft':
spaces.Enum('none', 'wooden_axe', 'wooden_pickaxe', 'stone_axe',
'stone_pickaxe', 'iron_axe', 'iron_pickaxe',
'furnace'),
'nearbySmelt':
spaces.Enum('none', 'iron_ingot', 'coal'),
'place':
spaces.Enum('none', 'dirt', 'stone', 'cobblestone',
'crafting_table', 'furnace', 'torch'),
'right':
spaces.Discrete(2),
'sneak':
spaces.Discrete(2),
'sprint':
spaces.Discrete(2)
}
self.observation_space = {
'equipped_items': {
'mainhand': {
'damage':
spaces.Box(np.full([2], -1),
np.full([2], 1),
dtype=np.int64),
'maxDamage':
spaces.Box(np.full([2], -1),
np.full([2], 1),
dtype=np.int64),
'type':
spaces.Enum('none', 'air', 'wooden_axe', 'wooden_pickaxe',
'stone_axe', 'stone_pickaxe', 'iron_axe',
'iron_pickaxe', 'other')
}
},
'inventory': {
'coal':
spaces.Box(np.full([1], -1), np.full([1], 1), dtype=np.int64),
'cobblestone':
spaces.Box(np.full([1], -1), np.full([1], 1), dtype=np.int64),
'crafting_table':
spaces.Box(np.full([1], -1), np.full([1], 1), dtype=np.int64),
'dirt':
spaces.Box(np.full([1], -1), np.full([1], 1), dtype=np.int64),
'furnace':
spaces.Box(np.full([1], -1), np.full([1], 1), dtype=np.int64),
'iron_axe':
spaces.Box(np.full([1], -1), np.full([1], 1), dtype=np.int64),
'iron_ingot':
spaces.Box(np.full([1], -1), np.full([1], 1), dtype=np.int64),
'iron_ore':
spaces.Box(np.full([1], -1), np.full([1], 1), dtype=np.int64),
'iron_pickaxe':
spaces.Box(np.full([1], -1), np.full([1], 1), dtype=np.int64),
'log':
spaces.Box(np.full([1], -1), np.full([1], -1), dtype=np.int64),
'planks':
spaces.Box(np.full([1], -1), np.full([1], -1), dtype=np.int64),
'stick':
spaces.Box(np.full([1], -1), np.full([1], -1), dtype=np.int64),
'stone':
spaces.Box(np.full([1], -1), np.full([1], -1), dtype=np.int64),
'stone_axe':
spaces.Box(np.full([1], -1), np.full([1], -1), dtype=np.int64),
'stone_pickaxe':
spaces.Box(np.full([1], -1), np.full([1], -1), dtype=np.int64),
'torch':
spaces.Box(np.full([1], -1), np.full([1], -1), dtype=np.int64),
'wooden_axe':
spaces.Box(np.full([1], -1), np.full([1], -1), dtype=np.int64),
'wooden_pickaxe':
spaces.Box(np.full([1], -1), np.full([1], -1), dtype=np.int64)
},
'pov':
spaces.Box(np.full([64, 64, 3], -1),
np.full([64, 64, 3], -1),
dtype=np.uint8)
}
self.reward_range = (-np.inf, np.inf)
self.metadata = {'render.modes': ['rgb_array', 'human']}
self.t = 0