def _add_symbols(self, obs):
obs = super()._add_symbols(obs)
obs["sqare_coords"] = self.metadata["square_coords"]
obs["ball_coords"] = self.metadata["ball_coords"]
obs["carrying"] = self.carrying
obs["ball_placed_correctly"] = self.has_ever_been_good
return obs
def _is_ball_placed_correctly(self):
object_goal = self.grid.get(*self.metadata["square_coords"])
return True if object_goal and object_goal.type == "ball" else False
def _handle_constraint_violation(self, done):
constraint_violation = 0
if done:
object_goal = self.grid.get(*self.metadata["square_coords"])
if not object_goal or object_goal.type != "ball":
constraint_violation += 1
self.metadata[
CONSTRAINT_VIOLATION_COUNT] += constraint_violation
return constraint_violation
def reset(self):
obs = super().reset()
# self.metadata["ball_coords"] = (1,3)
return obs
register(
id='MiniGrid-Ethical5x5Object-v0',
entry_point='ethical_rl.environments.five_by_five_object:Ethical5x5Object')
class MultiRoomEnvN6(MultiRoomEnv):
def __init__(self):
super().__init__(minNumRooms=6, maxNumRooms=6)
# class MultiRoomEnvN4r(MultiRoomEnv):
# def __init__(self):
# super().__init__(
# minNumRooms=1,
# maxNumRooms=5,
# maxRoomSize=7,
# grid_size=15
# )
register(id='MiniGrid-MultiRoom-N2-S4-v0',
entry_point='gym_minigrid.envs:MultiRoomEnvN2S4',
reward_threshold=1000.0)
register(id='MiniGrid-MultiRoom-N6-v0',
entry_point='gym_minigrid.envs:MultiRoomEnvN6',
reward_threshold=1000.0)
# Own environments:
register(id='MiniGrid-MultiRoom-N5r-v0',
entry_point='gym_minigrid.envs:MultiRoomEnv',
reward_threshold=1000.0,
kwargs={
'minNumRooms': 1,
'maxNumRooms': 5,
'maxRoomSize': 7,
'grid_size': 15
self.grid.set(*self._agent_default_pos, None)
self.agent_dir = self._rand_int(
0, 4) # assuming random start direction
else:
flag = False
for a in range(50):
if flag:
break
for b in range(50):
if (a, b) not in self.wallss:
self.place_agent((a, b))
flag = True
break
if self._goal_default_pos is not None:
goal = Goal()
self.put_obj(goal, *self._goal_default_pos)
goal.init_pos, goal.cur_pos = self._goal_default_pos
else:
self.place_obj(Goal())
self.mission = 'Reach the goal'
def step(self, action):
obs, reward, done, info = MiniGridEnv.step(self, action)
return obs, reward, done, info
register(id='MiniGrid-RandomMazeLarge-v0',
entry_point='gym_minigrid.envs:RandomMazeLarge')
class MultiRoomEnvN6(MultiRoomEnv):
def __init__(self):
super().__init__(minNumRooms=6, maxNumRooms=6)
# class MultiRoomEnvN4r(MultiRoomEnv):
# def __init__(self):
# super().__init__(
# minNumRooms=1,
# maxNumRooms=5,
# maxRoomSize=7,
# grid_size=15
# )
register(id='MiniGrid-MultiRoom-N2-S4-v0',
entry_point='gym_minigrid.envs:MultiRoomEnvN2S4',
reward_threshold=1000.0)
register(id='MiniGrid-MultiRoom-N6-v0',
entry_point='gym_minigrid.envs:MultiRoomEnvN6',
reward_threshold=1000.0)
# Own environments:
register(id='MiniGrid-MultiRoom-N2r-v0',
entry_point='gym_minigrid.envs:MultiRoomEnv',
reward_threshold=1000.0,
kwargs={
'minNumRooms': 1,
'maxNumRooms': 2,
'maxRoomSize': 6,
# Generate the surrounding walls
self.grid.wall_rect(0, 0, width, height)
# Place the agent in the top-left corner
self.start_pos = (1, 1)
self.start_dir = 0
# Place a goal square in the bottom-right corner
self.grid.set(width - 2, height - 2, Goal())
self.mission = "get to the green goal square"
class EmptyEnv6x6(EmptyEnv):
def __init__(self):
super().__init__(size=6)
class EmptyEnv16x16(EmptyEnv):
def __init__(self):
super().__init__(size=16)
register(id='MiniGrid-Empty-6x6-v0',
entry_point='gym_minigrid.envs:EmptyEnv6x6')
register(id='MiniGrid-Empty-8x8-v0', entry_point='gym_minigrid.envs:EmptyEnv')
register(id='MiniGrid-Empty-16x16-v0',
entry_point='gym_minigrid.envs:EmptyEnv16x16')
obs, reward, done, info = MiniGridEnv.step(self, action)
ax, ay = self.agentPos
tx, ty = self.targetPos
# Toggle/pickup action terminates the episode
if action == self.actions.toggle:
done = True
# Reward performing the wait action next to the target object
if action == self.actions.wait:
if abs(ax - tx) <= 1 and abs(ay - ty) <= 1:
reward = 1
done = True
return obs, reward, done, info
class GotoEnv8x8N2(GoToObjectEnv):
def __init__(self):
super().__init__(size=8, numObjs=2)
register(
id='MiniGrid-GoToObject-6x6-N2-v0',
entry_point='gym_minigrid.envs:GoToObjectEnv'
)
register(
id='MiniGrid-GoToObject-8x8-N2-v0',
entry_point='gym_minigrid.envs:GotoEnv8x8N2'
)
class EmptyEnvNxN(EmptyEnv):
def __init__(self, size=None, start=None, goal=None, four_corners=False):
'''
if _agent_start_pos == None:
_agent_start_pos_ = ((_size+1)/2 , (_size+1)/2)
if _goal_pos=None:
_goal_pos = (_size - 2, _size - 2)
'''
super().__init__(size=size + 2,
agent_start_pos=start,
goal_pos=goal,
four_corners=four_corners)
register(id='MiniGrid-Empty-NxN-v0',
entry_point='gym_minigrid.envs:EmptyEnvNxN')
register(id='MiniGrid-Empty-5x5-v0',
entry_point='gym_minigrid.envs:EmptyEnv5x5')
register(id='MiniGrid-Empty-Random-5x5-v0',
entry_point='gym_minigrid.envs:EmptyRandomEnv5x5')
register(id='MiniGrid-Empty-6x6-v0',
entry_point='gym_minigrid.envs:EmptyEnv6x6')
register(id='MiniGrid-Empty-Random-6x6-v0',
entry_point='gym_minigrid.envs:EmptyRandomEnv6x6')
register(id='MiniGrid-Empty-8x8-v0', entry_point='gym_minigrid.envs:EmptyEnv')
register(id='MiniGrid-Empty-16x16-v0',
ox, oy = (self.agent_pos[0] + u, self.agent_pos[1] + v)
tx, ty = self.target_pos
# If we picked up the wrong object, terminate the episode
if action == self.actions.pickup and self.carrying:
if self.carrying.type != self.move_type or self.carrying.color != self.moveColor:
done = True
# If successfully dropping an object near the target
if action == self.actions.drop and preCarrying:
if self.grid.get(ox, oy) is preCarrying:
if abs(ox - tx) <= 1 and abs(oy - ty) <= 1:
reward = self._reward()
done = True
return obs, reward, done, info
class PutNear8x8N3(PutNearEnv):
def __init__(self):
super().__init__(size=8, numObjs=3)
register(
id='MiniGrid-PutNear-6x6-N2-v0',
entry_point='gym_minigrid.envs:PutNearEnv'
)
register(
id='MiniGrid-PutNear-8x8-N3-v0',
entry_point='gym_minigrid.envs:PutNear8x8N3'
)
if success:
break
return True
class MultiRoomEnvN2S4(MultiRoomEnv):
def __init__(self):
super().__init__(minNumRooms=2, maxNumRooms=2, maxRoomSize=4)
class MultiRoomEnvN4S5(MultiRoomEnv):
def __init__(self):
super().__init__(minNumRooms=4, maxNumRooms=4, maxRoomSize=5)
class MultiRoomEnvN6(MultiRoomEnv):
def __init__(self):
super().__init__(minNumRooms=6, maxNumRooms=6)
register(id='MiniGrid-MultiRoom-N2-S4-v0',
entry_point='gym_minigrid.envs:MultiRoomEnvN2S4')
register(id='MiniGrid-MultiRoom-N4-S5-v0',
entry_point='gym_minigrid.envs:MultiRoomEnvN4S5')
register(id='MiniGrid-MultiRoom-N6-v0',
entry_point='gym_minigrid.envs:MultiRoomEnvN6')
# Select a random room to contain the key
while True:
keyRoom = self._rand_elem(self.rooms)
if keyRoom != lockedRoom:
break
keyPos = keyRoom.rand_pos(self)
self.grid.set(*keyPos, Key(lockedRoom.color))
# Randomize the player start position and orientation
self.start_pos = self.place_agent(
top=(lWallIdx, 0),
size=(rWallIdx-lWallIdx, height)
)
# Generate the mission string
self.mission = (
'get the %s key from the %s room, '
'then use it to unlock the %s door '
'so you can get to the goal'
) % (lockedRoom.color, keyRoom.color, lockedRoom.color)
def step(self, action):
obs, reward, done, info = MiniGridEnv.step(self, action)
return obs, reward, done, info
register(
id='MiniGrid-LockedRoom-v0',
entry_point='gym_minigrid.envs:LockedRoom'
)
break
return True
class MultiRoomEnvN2S4(MultiRoomEnv):
def __init__(self):
super().__init__(
minNumRooms=2,
maxNumRooms=2,
maxRoomSize=4
)
class MultiRoomEnvN6(MultiRoomEnv):
def __init__(self):
super().__init__(
minNumRooms=6,
maxNumRooms=6
)
register(
id='MiniGrid-MultiRoom-N2-S4-v0',
entry_point='gym_minigrid.envs:MultiRoomEnvN2S4',
reward_threshold=1000.0
)
register(
id='MiniGrid-MultiRoom-N6-v0',
entry_point='gym_minigrid.envs:MultiRoomEnvN6',
reward_threshold=1000.0
)
# Reward performing done action in front of the target door
if action == self.actions.done:
if (ax == tx and abs(ay - ty) == 1) or (ay == ty and abs(ax - tx) == 1):
reward = self._reward()
done = True
return obs, reward, done, info
class GoToDoor8x8Env(GoToDoorEnv):
def __init__(self):
super().__init__(size=8)
class GoToDoor6x6Env(GoToDoorEnv):
def __init__(self):
super().__init__(size=6)
register(
id='MiniGrid-GoToDoor-5x5-v0',
entry_point='gym_minigrid.envs:GoToDoorEnv'
)
register(
id='MiniGrid-GoToDoor-6x6-v0',
entry_point='gym_minigrid.envs:GoToDoor6x6Env'
)
register(
id='MiniGrid-GoToDoor-8x8-v0',
entry_point='gym_minigrid.envs:GoToDoor8x8Env'
)
self.mission = "use the key to open the door and then get to the goal"
class DoorKeyEnv5x5(DoorKeyEnv):
def __init__(self):
super().__init__(size=5)
class DoorKeyEnv6x6(DoorKeyEnv):
def __init__(self):
super().__init__(size=6)
class DoorKeyEnv16x16(DoorKeyEnv):
def __init__(self):
super().__init__(size=16)
register(
id='MiniGrid-DoorKey-5x5-v0',
entry_point='gym_minigrid.envs:DoorKeyEnv5x5'
)
register(
id='MiniGrid-DoorKey-6x6-v0',
entry_point='gym_minigrid.envs:DoorKeyEnv6x6'
)
register(
id='MiniGrid-DoorKey-8x8-v0',
entry_point='gym_minigrid.envs:DoorKeyEnv'
)
register(
id='MiniGrid-DoorKey-16x16-v0',
entry_point='gym_minigrid.envs:DoorKeyEnv16x16'
self.start_pos = (1, 1)
self.start_dir = 0
# Place a goal square in the bottom-right corner
self.grid.set(width - 2, height - 2, Goal())
self.mission = "get to the green goal square"
class EmptyEnv6x6(EmptyEnv):
def __init__(self):
super().__init__(size=6)
class EmptyEnv16x16(EmptyEnv):
def __init__(self):
super().__init__(size=16)
register(
id='MiniGrid-Empty-6x6-v0',
entry_point='gym_minigrid.envs:EmptyEnv6x6'
)
register(
id='MiniGrid-Empty-8x8-v0',
entry_point='gym_minigrid.envs:EmptyEnv'
)
register(
id='MiniGrid-Empty-16x16-v0',
entry_point='gym_minigrid.envs:EmptyEnv16x16'
)
self._t = 0
self.state = self.np_random.rand() * 0.5
def get_state(self):
return self.state
def get_signature(self):
sig = {'agent_pos': np.copy(self.state), 't': copy(self._t)}
return sig
def set_signature(self, sig):
self.state = np.copy(sig['agent_pos'])
self._t = copy(sig['t'])
register(id='MiniGrid-RiverSwim-continuous-v0',
entry_point='envs.river_swim_continuous:RiverSwimContinuous')
if __name__ == '__main__':
fail = [i * 0.1 for i in range(10)]
for f in fail:
mdp = RiverSwimContinuous(horizon=40, dim=10, fail=f)
gamma = 0.99
num_episodes = 1000
rets = []
for i in range(num_episodes):
ret = 0
t = 0
done = False
s = mdp.reset()
while not done:
a = 1
def __init__(self):
super().__init__(minNumRooms=6, maxNumRooms=6)
class MultiRoomEnvN2S8_noterm_po(MultiRoomEnv):
def __init__(self):
super().__init__(minNumRooms=2,
maxNumRooms=2,
maxRoomSize=8,
terminate_at_goal=False,
max_steps=1000,
partial_obs=True)
register(id='MiniGrid-MultiRoom-N2-S4-v0',
entry_point='gym_minigrid.envs:MultiRoomEnvN2S4',
reward_threshold=1000.0)
register(id='MiniGrid-MultiRoom-N2-S8-v0',
entry_point='gym_minigrid.envs:MultiRoomEnvN2S8',
reward_threshold=1000.0)
register(id='MiniGrid-MultiRoom-N2-S8-v4',
entry_point='gym_minigrid.envs:MultiRoomEnvN2S8_po',
reward_threshold=1000.0)
register(id='MiniGrid-MultiRoom-N2-S8-NT-v0',
entry_point='gym_minigrid.envs:MultiRoomEnvN2S8_noterm',
reward_threshold=1000.0)
register(id='MiniGrid-MultiRoom-N2-S8-NT-v4',
entry_point='gym_minigrid.envs:MultiRoomEnvN2S8_noterm_po',
reward_threshold=1000.0)
register(id='MiniGrid-MultiRoom-N6-v0',
entry_point='gym_minigrid.envs:MultiRoomEnvN6',
reward = self._reward()
done = True
return obs, reward, done, info
class GoToDoor8x8Env(GoToDoorEnv):
def __init__(self):
super().__init__(size=8)
class GoToDoor6x6Env(GoToDoorEnv):
def __init__(self):
super().__init__(size=6)
register(
id='MiniGrid-GoToDoor-5x5-v0',
entry_point='gym_minigrid.envs:GoToDoorEnv'
)
register(
id='MiniGrid-GoToDoor-6x6-v0',
entry_point='gym_minigrid.envs:GoToDoor6x6Env'
)
register(
id='MiniGrid-GoToDoor-8x8-v0',
entry_point='gym_minigrid.envs:GoToDoor8x8Env'
)
class DoorKeyEnv12x12(DoorKeyEnv):
def __init__(self):
super().__init__(size=12)
class DoorKeyEnv14x14(DoorKeyEnv):
def __init__(self):
super().__init__(size=14)
class DoorKeyEnv16x16(DoorKeyEnv):
def __init__(self):
super().__init__(size=16)
register(id='MiniGrid-DoorKey-5x5-v0',
entry_point='gym_minigrid.envs:DoorKeyEnv5x5')
register(id='MiniGrid-DoorKey-6x6-v0',
entry_point='gym_minigrid.envs:DoorKeyEnv6x6')
register(id='MiniGrid-DoorKey-8x8-v0',
entry_point='gym_minigrid.envs:DoorKeyEnv')
register(id='MiniGrid-DoorKey-16x16-v0',
entry_point='gym_minigrid.envs:DoorKeyEnv16x16')
register(id='MiniGrid-DoorKey-10x10-v0',
entry_point='gym_minigrid.envs:DoorKeyEnv10x10')
register(id='MiniGrid-DoorKey-12x12-v0',
entry_point='gym_minigrid.envs:DoorKeyEnv12x12')
color = self._rand_elem(COLOR_NAMES)
self.grid.set(*pos, Door(color))
# Randomize the player start position and orientation
self.place_agent()
# Place random objects in the world
types = ['key', 'ball', 'box']
for i in range(0, 12):
objType = self._rand_elem(types)
objColor = self._rand_elem(COLOR_NAMES)
if objType == 'key':
obj = Key(objColor)
elif objType == 'ball':
obj = Ball(objColor)
elif objType == 'box':
obj = Box(objColor)
self.place_obj(obj)
# No explicit mission in this environment
self.mission = ''
def step(self, action):
obs, reward, done, info = MiniGridEnv.step(self, action)
return obs, reward, done, info
register(
id='MiniGrid-Playground-v0',
entry_point='gym_minigrid.envs:PlaygroundV0'
)
#!/usr/bin/env python3
import time
import random
import argparse
import numpy as np
import gym
import gym_minigrid
from gym_minigrid.wrappers import *
from gym_minigrid.window import Window
from gym_minigrid.register import register
register(id='MiniGrid-Relational-Obstacles-10x10-v0',
entry_point='env:RelationalObstaclesEnv10x10')
parser = argparse.ArgumentParser()
parser.add_argument("--env",
help="gym environment to load",
default='MiniGrid-Relational-Obstacles-10x10-v0')
parser.add_argument("--seed",
type=int,
help="random seed to generate the environment with",
default=-1)
parser.add_argument("--tile_size",
type=int,
help="size at which to render tiles",
default=32)
parser.add_argument('--agent_view',
default=False,
help="draw the agent sees (partially observable view)",
action='store_true')
args = parser.parse_args()
if self.carrying and self.carrying.type == 'key' and done is True:
# delayed reward if carrying key at goal
reward += self.key_reward
return obs, reward, done, info
class KeyGoalEnv6x6(KeyGoalEnv):
def __init__(self):
super().__init__(size=6, max_steps=10 * 6**2)
class KeyGoalEnv8x8(KeyGoalEnv):
def __init__(self):
super().__init__(size=8, max_steps=10 * 8**2)
class KeyGoalEnv10x10(KeyGoalEnv):
def __init__(self):
super().__init__(size=10, max_steps=10 * 10**2)
register(id='MiniGrid-KeyGoal-6x6-v0',
entry_point='gym_minigrid.envs:KeyGoalEnv6x6')
register(id='MiniGrid-KeyGoal-8x8-v0',
entry_point='gym_minigrid.envs:KeyGoalEnv8x8')
register(id='MiniGrid-KeyGoal-10x10-v0',
entry_point='gym_minigrid.envs:KeyGoalEnv10x10')
help="gym environment to load",
default='hwft-5e')
parser.add_argument("--seed",
type=int,
help="random seed to generate the environment with",
default=-1)
parser.add_argument("--tile_size",
type=int,
help="size at which to render tiles",
default=32)
parser.add_argument('--agent_view',
default=False,
help="draw the agent sees (partially observable view)",
action='store_true')
args = parser.parse_args()
register(id=env_dict[args.env_name][0], entry_point=env_dict[args.env_name][1])
env = gym.make(env_dict[args.env_name][0])
if args.agent_view:
env = RGBImgPartialObsWrapper(env)
env = ImgObsWrapper(env)
window = Window('gym_minigrid - ' + args.env_name)
window.reg_key_handler(key_handler)
reset()
# Blocking event loop
window.show(block=True)
class EmptyEnv18x18(EmptyEnv):
def __init__(self, **kwargs):
super().__init__(size=18, **kwargs)
class EmptyEnv30x30(EmptyEnv):
def __init__(self, **kwargs):
super().__init__(size=30, **kwargs)
class EmptyEnv50x50(EmptyEnv):
def __init__(self, **kwargs):
super().__init__(size=50, **kwargs)
register(id='MiniGrid-EmptyCorridor-5x1-v0',
entry_point='gym_minigrid.envs:EmptyCorridor5x1')
register(id='MiniGrid-EmptyCorridor-10x1-v0',
entry_point='gym_minigrid.envs:EmptyCorridor10x1')
register(id='MiniGrid-Empty-5x5-v0',
entry_point='gym_minigrid.envs:EmptyEnv5x5')
register(id='MiniGrid-Empty-InverseGoal-5x5-v0',
entry_point='gym_minigrid.envs:EmptyEnv5x5InverseGoal')
register(id='MiniGrid-Empty-Random-5x5-v0',
entry_point='gym_minigrid.envs:EmptyRandomEnv5x5')
register(id='MiniGrid-Empty-6x6-v0',
entry_point='gym_minigrid.envs:EmptyEnv6x6')
# Set goal
assert self.goal_pos is not None, "Goal position is None. Call env.reset_task() before env.reset()"
self.put_obj(Goal(), self.goal_pos[0] + 1, self.goal_pos[1] + 1)
def _reward(self):
dist = np.linalg.norm(np.array(self.agent_pos) - self.goal_pos)
return -dist
def reset_task(self, task):
if task is None:
task = self.np_random.randint(0, self.size - 2, size=(2, ))
assert len(task) == 2, "Must in format of (col, row)"
assert task[0] >= 0 and task[0] < self.size - 2
assert task[1] >= 0 and task[1] < self.size - 2
self.goal_pos = np.array(task)
def sample_tasks(self, num_tasks):
tasks = self.np_random.randint(0, self.size - 2, size=(num_tasks, 2))
# tasks = self.np_random.randint(1, 2, size=(num_tasks, 2))
return tasks
class EmptyEnv5x5(EmptyEnv):
def __init__(self, **kwargs):
super().__init__(size=5, **kwargs)
register(id='MiniGrid-Empty-5x5-v0',
entry_point='gym_minigrid.envs:EmptyEnv5x5')
value=value,
color="blue")
self.mission = "find the highest reward"
def reset(self):
if self.fixed_seed is not None:
self.seed(1)
return super().reset()
register(
id="MiniGrid-MultipleRewardRoomS4R1-v0",
entry_point="gym_minigrid.envs:MultipleRewardRoom",
kwargs={
"room_size": 4,
"num_rows": 1,
"fixed_seed": 1
},
)
register(
id="MiniGrid-MultipleRewardRoomS4R2-v0",
entry_point="gym_minigrid.envs:MultipleRewardRoom",
kwargs={
"room_size": 4,
"num_rows": 2,
"fixed_seed": 1
},
)
class DynamicObstaclesEnv6x6(DynamicObstaclesEnv):
def __init__(self):
super().__init__(size=6, n_obstacles=3)
class DynamicObstaclesRandomEnv6x6(DynamicObstaclesEnv):
def __init__(self):
super().__init__(size=6, agent_start_pos=None, n_obstacles=3)
class DynamicObstaclesRandomEnv10x10(DynamicObstaclesEnv):
def __init__(self):
super().__init__(size=10, agent_start_pos=None, n_obstacles=5)
register(id='MiniGrid-Dynamic-Obstacles-5x5-v0',
entry_point='gym_minigrid.envs:DynamicObstaclesEnv5x5')
register(id='MiniGrid-Dynamic-Obstacles-Random-5x5-v0',
entry_point='gym_minigrid.envs:DynamicObstaclesRandomEnv5x5')
register(id='MiniGrid-Dynamic-Obstacles-6x6-v0',
entry_point='gym_minigrid.envs:DynamicObstaclesEnv6x6')
register(id='MiniGrid-Dynamic-Obstacles-Random-6x6-v0',
entry_point='gym_minigrid.envs:DynamicObstaclesRandomEnv6x6')
register(id='MiniGrid-Dynamic-Obstacles-8x8-v0',
entry_point='gym_minigrid.envs:DynamicObstaclesEnv')
register(id='MiniGrid-Dynamic-Obstacles-Random-10x10-v0',
entry_point='gym_minigrid.envs:DynamicObstaclesRandomEnv10x10')
maxRoomSize=5,
gridSize=14,
)
class MultiRoomEnvN4S7Random(MultiRoomEnv):
def __init__(self):
super().__init__(
minNumRooms=1,
maxNumRooms=4,
maxRoomSize=7,
gridSize=13,
)
register(id='MiniGrid-MultiRoom-N2-S4-v0',
entry_point='gym_minigrid.envs:MultiRoomEnvN2S4')
register(id='MiniGrid-MultiRoom-N4-S5-v0',
entry_point='gym_minigrid.envs:MultiRoomEnvN4S5')
register(id='MiniGrid-MultiRoom-N6-v0',
entry_point='gym_minigrid.envs:MultiRoomEnvN6')
register(id='MiniGrid-MultiRoom-N4-Random-v0',
entry_point='gym_minigrid.envs:MultiRoomEnvN4Random')
# Custom levels
register(id='MiniGrid-MultiRoom-N4-S7-Random-v0',
entry_point='gym_minigrid.envs:MultiRoomEnvN4S7Random')
register(id='MiniGrid-MultiRoom-N2-S4-G14-v0',
class FetchKeyEnv6x6N5(FetchKeyEnv):
def __init__(self):
super().__init__(size=6, numObjs=5)
class FetchKeyEnv8x8N5(FetchKeyEnv):
def __init__(self):
super().__init__(size=8, numObjs=5)
class FetchKeyEnv16x16N5(FetchKeyEnv):
def __init__(self):
super().__init__(size=16, numObjs=5)
register(id='MiniGrid-fetchKey-5x5-N5-v0',
entry_point='gym_minigrid.envs:FetchKeyEnv5x5N5')
register(id='MiniGrid-fetchKey-6x6-N5-v0',
entry_point='gym_minigrid.envs:FetchKeyEnv6x6N5')
register(id='MiniGrid-fetchKey-8x8-N5-v0',
entry_point='gym_minigrid.envs:FetchKeyEnv8x8N5')
register(id='MiniGrid-fetchKey-16x16-N5-v0',
entry_point='gym_minigrid.envs:FetchKeyEnv16x16N5')
#############################################################
class FetchKeyEnv5x5N3(FetchKeyEnv):
def __init__(self):
self.grid.set(width - 2, height - 2, Goal())
# Place water
self.grid.set(width - 5, height - 2, Water())
# Set start position
self.start_pos = (1, 1)
self.start_dir = 0
self.mission = "get to the green goal square without moving on water"
class UnsafeEnv6x6(UnsafeEnv):
def __init__(self):
super().__init__(size=6)
class UnsafeEnv16x16(UnsafeEnv):
def __init__(self):
super().__init__(size=16)
register(id='MiniGrid-Unsafe-6x6-v0',
entry_point='gym_minigrid.envs:UnsafeEnv6x6')
register(id='MiniGrid-Unsafe-8x8-v0',
entry_point='gym_minigrid.envs:UnsafeEnv')
register(id='MiniGrid-Unsafe-16x16-v0',
entry_point='gym_minigrid.envs:UnsafeEnv16x16')
num_cols=2,
room_size=room_size,
max_steps=8 * room_size**2,
seed=seed)
def _gen_grid(self, width, height):
super()._gen_grid(width, height)
# Make sure the two rooms are directly connected by a locked door
door, _ = self.add_door(0, 0, 0, locked=True)
# Add a key to unlock the door
self.add_object(0, 0, 'key', door.color)
self.place_agent(0, 0)
self.door = door
self.mission = "open the door"
def step(self, action):
obs, reward, done, info = super().step(action)
if action == self.actions.toggle:
if self.door.is_open:
reward = self._reward()
done = True
return obs, reward, done, info
register(id='MiniGrid-Unlock-v0', entry_point='gym_minigrid.envs:Unlock')
return obs, reward, done, info
def _reward(self):
"""
Reward received upon goal collection
"""
return 1
class CollectEnv9x9(CollectEnv):
def __init__(self):
super().__init__(size=9,
agent_start_pos=None,
n_goals=15,
num_crossings=0)
class CollectEnvStochastic9x9(CollectEnv):
def __init__(self):
super().__init__(size=9,
agent_start_pos=None,
n_goals=15,
num_crossings=0,
reward_noise=0.15)
register(id='MiniGrid-Collect-9x9-v0',
entry_point='gym_minigrid.envs:CollectEnv9x9')
register(id='MiniGrid-Collect-Stochastic-9x9-v0',
entry_point='gym_minigrid.envs:CollectEnvStochastic9x9')
obj, _ = self.add_object(1, 0, kind="box")
# Make sure the two rooms are directly connected by a locked door
door, pos = self.add_door(0, 0, 0, locked=True)
# Block the door with a ball
color = self._rand_color()
self.grid.set(pos[0] - 1, pos[1], Ball(color))
# Add a key to unlock the door
self.add_object(0, 0, 'key', door.color)
self.place_agent(0, 0)
self.obj = obj
self.mission = "pick up the %s %s" % (obj.color, obj.type)
def step(self, action):
obs, reward, done, info = super().step(action)
if action == self.actions.pickup:
if self.carrying and self.carrying == self.obj:
reward = self._reward()
done = True
# print("DONE: ", done)
if done:
self.count += 1
print("COUNT: ", self.count)
return obs, reward, done, info
register(id='MiniGrid-BlockedUnlockPickup-v0',
entry_point='gym_minigrid.envs:BlockedUnlockPickup')
class EmptyGridEnvOmniNoGoal6x6(EmptyEnvNoGoal):
def __init__(self):
super().__init__(size=6, mode='grid', action_mode='omni')
class EmptyGridEnvOmniNoGoal8x8(EmptyEnvNoGoal):
def __init__(self):
super().__init__(size=8, mode='grid', action_mode='omni')
class EmptyGridEnvOmniNoGoal16x16(EmptyEnvNoGoal):
def __init__(self):
super().__init__(size=16, mode='grid', action_mode='omni')
register(id='MiniGrid-Empty-NoGoal-6x6-v0',
entry_point='gym_minigrid.envs:EmptyEnvNoGoal6x6')
register(id='MiniGrid-Empty-NoGoal-8x8-v0',
entry_point='gym_minigrid.envs:EmptyEnvNoGoal')
register(id='MiniGrid-Empty-NoGoal-16x16-v0',
entry_point='gym_minigrid.envs:EmptyEnvNoGoal16x16')
register(id='MiniGrid-Empty-NoGoal-6x6-v1',
entry_point='gym_minigrid.envs:EmptyGridEnvNoGoal6x6')
register(id='MiniGrid-Empty-NoGoal-8x8-v1',
entry_point='gym_minigrid.envs:EmptyGridEnvNoGoal')
register(id='MiniGrid-Empty-NoGoal-16x16-v1',
entry_point='gym_minigrid.envs:EmptyGridEnvNoGoal16x16')
if self.carrying:
if self.carrying.color == self.targetColor and \
self.carrying.type == self.targetType:
reward = 1
done = True
else:
reward = 0
done = True
return obs, reward, done, info
class FetchEnv5x5N2(FetchEnv):
def __init__(self):
super().__init__(size=5, numObjs=2)
class FetchEnv6x6N2(FetchEnv):
def __init__(self):
super().__init__(size=6, numObjs=2)
register(id='MiniGrid-Fetch-5x5-N2-v0',
entry_point='gym_minigrid.envs:FetchEnv5x5N2')
register(id='MiniGrid-Fetch-6x6-N2-v0',
entry_point='gym_minigrid.envs:FetchEnv6x6N2')
register(id='MiniGrid-Fetch-8x8-N3-v0',
entry_point='gym_minigrid.envs:FetchEnv')
if j + 1 < 2:
self.grid.horz_wall(xL, yB, room_w)
pos = (self._rand_int(xL + 1, xR), yB)
self.grid.set(*pos, None)
# Randomize the player start position and orientation
if self._agent_default_pos is not None:
self.agent_pos = self._agent_default_pos
self.grid.set(*self._agent_default_pos, None)
self.agent_dir = self._rand_int(0, 4) # assuming random start direction
else:
self.place_agent()
if self._goal_default_pos is not None:
goal = Goal()
self.put_obj(goal, *self._goal_default_pos)
goal.init_pos, goal.cur_pos = self._goal_default_pos
else:
self.place_obj(Goal())
self.mission = 'Reach the goal'
def step(self, action):
obs, reward, done, info = MiniGridEnv.step(self, action)
return obs, reward, done, info
register(
id='MiniGrid-FourRooms-v0',
entry_point='gym_minigrid.envs:FourRoomsEnv'
)
self.mission = "use the key to open the door and then get to the goal"
class DoorKeyEnv5x5(DoorKeyEnv):
def __init__(self):
super().__init__(size=5)
class DoorKeyEnv6x6(DoorKeyEnv):
def __init__(self):
super().__init__(size=6)
class DoorKeyEnv16x16(DoorKeyEnv):
def __init__(self):
super().__init__(size=16)
register(
id='MiniGrid-DoorKey-5x5-v0',
entry_point='gym_minigrid.envs:DoorKeyEnv5x5'
)
register(
id='MiniGrid-DoorKey-6x6-v0',
entry_point='gym_minigrid.envs:DoorKeyEnv6x6'
)
register(
id='MiniGrid-DoorKey-8x8-v0',
entry_point='gym_minigrid.envs:DoorKeyEnv'
)
register(
id='MiniGrid-DoorKey-16x16-v0',
entry_point='gym_minigrid.envs:DoorKeyEnv16x16'
# Generate the wall which separate the rooms
self.grid.vert_wall(3, 0, 5)
self.grid.horz_wall(0, 4, 5)
# Place the door which separate the rooms
self.grid.set(3, 2, Door(self._rand_elem(sorted(set(COLOR_NAMES)))))
# Place a goal square in the bottom-right corner
self.grid.set(4, 2, Goal())
# Add the rooms
self.roomList = []
self.roomList.append(Room(0, (width / 2 - 1, height), (0, 0), True))
self.roomList.append(Room(1, (width, height), (width / 2, 0), False))
# Set room entry and exit that are needed
self.roomList[1].setEntryDoor((3, 2))
self.roomList[0].setExitDoor((3, 2))
# Add the switch of the second room in the first one
switchRoom2 = LightSwitch()
switchRoom2.affectRoom(self.roomList[1])
self.grid.set(2, 3, switchRoom2)
self.switchPosition = []
self.switchPosition.append((2, 3))
self.mission = "get to the green goal square without moving on water"
register(id='MiniGrid-DoorEnv-10x10-v0',
entry_point='gym_minigrid.envs:DoorEnv')
blue_door_opened_after = self.blue_door.is_open
if blue_door_opened_after:
if red_door_opened_before:
reward = self._reward()
done = True
else:
reward = 0
done = True
elif red_door_opened_after:
if blue_door_opened_before:
reward = 0
done = True
return obs, reward, done, info
class RedBlueDoorEnv6x6(RedBlueDoorEnv):
def __init__(self):
super().__init__(size=6)
register(
id='MiniGrid-RedBlueDoors-6x6-v0',
entry_point='gym_minigrid.envs:RedBlueDoorEnv6x6'
)
register(
id='MiniGrid-RedBlueDoors-8x8-v0',
entry_point='gym_minigrid.envs:RedBlueDoorEnv'
)
