def env_ctor():
return suite_gym.wrap_env(DeepmindLabEnv(scene=scene,
action_repeat=frame_skip),
discount=discount,
max_episode_steps=max_episode_steps,
gym_env_wrappers=gym_env_wrappers,
env_wrappers=env_wrappers)
def env_ctor(env_id=None):
return suite_gym.wrap_env(env,
env_id=env_id,
discount=discount,
max_episode_steps=max_episode_steps,
gym_env_wrappers=gym_env_wrappers,
alf_env_wrappers=alf_env_wrappers)
def env_ctor(port):
gym_env = gym_spec.make(port=port)
return suite_gym.wrap_env(gym_env,
discount=discount,
max_episode_steps=max_episode_steps,
gym_env_wrappers=gym_env_wrappers,
env_wrappers=env_wrappers,
spec_dtype_map=spec_dtype_map)
def load(environment_name,
env_id=None,
discount=1.0,
max_episode_steps=None,
unconstrained=False,
gym_env_wrappers=(),
alf_env_wrappers=()):
"""Loads the selected environment and wraps it with the specified wrappers.
Note that by default a ``TimeLimit`` wrapper is used to limit episode lengths
to the default benchmarks defined by the registered environments.
Args:
environment_name: Name for the environment to load.
env_id: A scalar ``Tensor`` of the environment ID of the time step.
discount: Discount to use for the environment.
max_episode_steps: If None or 0 the ``max_episode_steps`` will be set to
the default step limit -1 defined in the environment. Otherwise
``max_episode_steps`` will be set to the smaller value of the two.
unconstrained (bool): if True, the suite will be used just as an
unconstrained environment. The reward will always be scalar without
including constraints.
gym_env_wrappers: Iterable with references to wrapper classes to use
directly on the gym environment.
alf_env_wrappers: Iterable with references to wrapper classes to use on
the torch environment.
Returns:
An AlfEnvironment instance.
"""
# We can directly make the env here because none of the safety gym tasks
# is registered with a ``max_episode_steps`` argument (the
# ``gym.wrappers.time_limit.TimeLimit`` won't be applied). But each task
# will inherently manage the time limit through ``env.num_steps``.
env = gym.make(environment_name)
# fill all env info with default values
env = CompleteEnvInfo(env, environment_name)
# make vector reward
if not unconstrained:
env = VectorReward(env)
# Have to -1 on top of the original env max steps here, because the
# underlying gym env will output ``done=True`` when reaching the time limit
# ``env.num_steps`` (before the ``AlfGymWrapper``), which is incorrect:
# https://github.com/openai/safety-gym/blob/f31042f2f9ee61b9034dd6a416955972911544f5/safety_gym/envs/engine.py#L1302
if not max_episode_steps: # None or 0
max_episode_steps = env.num_steps - 1
max_episode_steps = min(env.num_steps - 1, max_episode_steps)
return suite_gym.wrap_env(env,
env_id=env_id,
discount=discount,
max_episode_steps=max_episode_steps,
gym_env_wrappers=gym_env_wrappers,
alf_env_wrappers=alf_env_wrappers)
def env_ctor(env_id=None):
return suite_gym.wrap_env(
env,
env_id=env_id,
discount=discount,
max_episode_steps=max_episode_steps,
gym_env_wrappers=gym_env_wrappers,
alf_env_wrappers=alf_env_wrappers,
image_channel_first=False)
def load(game,
env_args=dict(),
discount=1.0,
frame_skip=None,
frame_stack=None,
gym_env_wrappers=(),
env_wrappers=(),
max_episode_steps=0,
spec_dtype_map=None):
"""Loads the specified simple game and wraps it.
Args:
game (str): name for the environment to load. The game should have been
defined in the sub-directory './simple/'.
env_args (dict): extra args for creating the game.
discount (float): discount to use for the environment.
frame_skip (int): the time interval at which the agent experiences the
game.
frame_stack (int): stack so many latest frames as the observation input.
gym_env_wrappers (list): list of gym env wrappers
env_wrappers (list): list of tf_agents env wrappers
max_episode_steps (int): max number of steps for an episode.
spec_dtype_map (dict): a dict that maps gym specs to tf dtypes to use as
the default dtype for the tensors. An easy way how to configure a
custom mapping through Gin is to define a gin-configurable function
that returns desired mapping and call it in your Gin config file, for
example: `suite_socialbot.load.spec_dtype_map = @get_custom_mapping()`.
Returns:
A PyEnvironmentBase instance.
"""
if spec_dtype_map is None:
spec_dtype_map = {gym.spaces.Box: np.float32}
if game == "NoisyArray":
env = NoisyArray(**env_args)
else:
assert False, "No such simple environment!"
if frame_skip:
env = FrameSkip(env, frame_skip)
if frame_stack:
env = FrameStack(env, stack_size=frame_stack)
return suite_gym.wrap_env(env,
discount=discount,
max_episode_steps=max_episode_steps,
gym_env_wrappers=gym_env_wrappers,
env_wrappers=env_wrappers,
spec_dtype_map=spec_dtype_map,
auto_reset=True)
def env_ctor(env_id=None):
env_args = [game, state] if state else [game]
env = retro.make(*env_args, record=record)
buttons = env.buttons
env = MarioXReward(env)
if frame_skip:
env = FrameSkip(env, frame_skip)
env = ProcessFrame84(env, crop=crop)
if frame_stack:
env = FrameStack(env, stack_size=frame_stack)
env = FrameFormat(env, data_format=data_format)
env = LimitedDiscreteActions(env, buttons)
return suite_gym.wrap_env(
env,
env_id=env_id,
discount=discount,
max_episode_steps=max_episode_steps,
gym_env_wrappers=gym_env_wrappers,
alf_env_wrappers=alf_env_wrappers,
auto_reset=True)
def load(game,
env_args=dict(),
discount=1.0,
frame_skip=None,
frame_stack=None,
gym_env_wrappers=(),
alf_env_wrappers=(),
max_episode_steps=0):
"""Loads the specified simple game and wraps it.
Args:
game (str): name for the environment to load. The game should have been
defined in the sub-directory ``./simple/``.
env_args (dict): extra args for creating the game.
discount (float): discount to use for the environment.
frame_skip (int): the time interval at which the agent experiences the
game.
frame_stack (int): stack so many latest frames as the observation input.
gym_env_wrappers (list): list of gym env wrappers.
alf_env_wrappers (list): list of ALF env wrappers.
max_episode_steps (int): max number of steps for an episode.
Returns:
An AlfEnvironment instance.
"""
if game == "NoisyArray":
env = NoisyArray(**env_args)
else:
assert False, "No such simple environment!"
if frame_skip:
env = FrameSkip(env, frame_skip)
if frame_stack:
env = FrameStack(env, stack_size=frame_stack)
return suite_gym.wrap_env(
env,
discount=discount,
max_episode_steps=max_episode_steps,
gym_env_wrappers=gym_env_wrappers,
alf_env_wrappers=alf_env_wrappers,
auto_reset=True)
def load(environment_name,
env_id=None,
discount=1.0,
max_episode_steps=None,
gym_env_wrappers=(),
alf_env_wrappers=(),
env_config=None):
"""Loads the selected environment and wraps it with the specified wrappers.
Note that by default a TimeLimit wrapper is used to limit episode lengths
to the default benchmarks defined by the registered environments.
Args:
environment_name (str): Name for the environment to load.
env_id (int): (optional) ID of the environment.
discount (float): Discount to use for the environment.
max_episode_steps (int): If None or 0 the ``max_episode_steps`` will be
set to the default step limit defined in the environment. Otherwise
``max_episode_steps`` will be set to the smaller value of the two.
gym_env_wrappers (Iterable): Iterable with references to gym_wrappers
classes to use directly on the gym environment.
alf_env_wrappers (Iterable): Iterable with references to alf_wrappers
classes to use on the ALF environment.
env_config (dict|None): a dictionary for configuring some aspects of the
environment. If is None, the default configuration will be used.
Please refer to the ``default_env_config`` below for
an example config and the doc for more details:
https://highway-env.readthedocs.io/en/latest/user_guide.html
Returns:
An AlfEnvironment instance.
"""
assert environment_name in {
"highway-v0", "merge-v0", "roundabout-v0", "intersection-v0",
"parking-v0"
}, "wrong highway environment name"
gym_spec = gym.spec(environment_name)
gym_env = gym_spec.make()
if env_config is None:
default_env_config = {
"observation": {
"type":
"Kinematics",
"vehicles_count":
5,
"features": [
"presence", "x", "y", "vx", "vy", "cos_h", "sin_h"
],
"features_range": {
"x": [-100, 100],
"y": [-100, 100],
"vx": [-20, 20],
"vy": [-20, 20]
},
"absolute":
False,
"order":
"sorted"
},
"action": {
"type": "ContinuousAction"
}
}
env_config = default_env_config
gym_env.configure(env_config)
gym_env.reset()
# currently flatten the observations, will support other ways later
gym_env = FlattenObservation(gym_env)
gym_env = RemoveActionEnvInfo(gym_env)
gym_env = ActionScalarization(gym_env)
# In the original environment, the last step due to time limit is not
# differentiated from those due to other reasons (e.g. crash):
# https://github.com/eleurent/highway-env/blob/ede285567a164a58b5bf8a78f1a6792f5a13a3fb/highway_env/envs/highway_env.py#L97-L99
# Here we -1 on top of the max steps specified by config["duration"] and
# use the time_limit_wrapper from alf to handle the last step correctly.
if not max_episode_steps:
max_episode_steps = gym_env.config["duration"] - 1
max_episode_steps = min(gym_env.config["duration"] - 1, max_episode_steps)
return suite_gym.wrap_env(
gym_env,
env_id=env_id,
discount=discount,
max_episode_steps=max_episode_steps,
gym_env_wrappers=gym_env_wrappers,
alf_env_wrappers=alf_env_wrappers)