def run_linear(time_limit=_DEFAULT_TIME_LIMIT, random=None, environment_kwargs=None):
"""Returns the run task."""
physics = Physics.from_xml_string(*get_model_and_assets())
task = Cheetah(random=random, distractor_style=1)
environment_kwargs = environment_kwargs or {}
return control.Environment(physics, task, time_limit=time_limit,
**environment_kwargs)
def hard(time_limit=_DEFAULT_TIME_LIMIT, random=None, environment_kwargs=None):
"""Returns the hard point_mass task."""
physics = Physics.from_xml_string(*get_model_and_assets())
task = PointMass(randomize_gains=True, random=random)
environment_kwargs = environment_kwargs or {}
return control.Environment(
physics, task, time_limit=time_limit, **environment_kwargs)
def run(time_limit=_DEFAULT_TIME_LIMIT, random=None, environment_kwargs=None, setting_kwargs=None):
"""Returns the run task."""
physics = Physics.from_xml_string(*common.settings.get_model_and_assets_from_setting_kwargs('cheetah.xml', setting_kwargs))
task = Cheetah(random=random)
environment_kwargs = environment_kwargs or {}
return control.Environment(physics, task, time_limit=time_limit,
**environment_kwargs)
def _make_lqr(n_bodies, n_actuators, control_cost_coef, time_limit, random,
environment_kwargs):
"""Returns a LQR environment.
Args:
n_bodies: An int, number of bodies of the LQR.
n_actuators: An int, number of actuated bodies of the LQR. `n_actuators`
should be less or equal than `n_bodies`.
control_cost_coef: A number, the coefficient of the control cost.
time_limit: An int, maximum time for each episode in seconds.
random: Either an existing `numpy.random.RandomState` instance, an
integer seed for creating a new `RandomState`, or None to select a seed
automatically.
environment_kwargs: A `dict` specifying keyword arguments for the
environment, or None.
Returns:
A LQR environment with `n_bodies` bodies of which first `n_actuators` are
actuated.
"""
if not isinstance(random, np.random.RandomState):
random = np.random.RandomState(random)
model_string, assets = get_model_and_assets(n_bodies,
n_actuators,
random=random)
physics = Physics.from_xml_string(model_string, assets=assets)
task = LQRLevel(control_cost_coef, random=random)
environment_kwargs = environment_kwargs or {}
return control.Environment(physics,
task,
time_limit=time_limit,
**environment_kwargs)
def hard(time_limit=_DEFAULT_TIME_LIMIT, random=None, environment_kwargs=None):
"""Returns reacher with sparse reward with 1e-2 tol and randomized target."""
physics = Physics.from_xml_string(*get_model_and_assets())
task = Reacher(target_size=_SMALL_TARGET, random=random)
environment_kwargs = environment_kwargs or {}
return control.Environment(
physics, task, time_limit=time_limit, **environment_kwargs)
def swingup(time_limit=_DEFAULT_TIME_LIMIT, random=None,
environment_kwargs=None, c_m=1., p_m=.1, p_l=1., p_r=.045, dt=.1):
"""Returns the Cartpole Swing-Up task."""
physics = Physics.from_xml_string(*get_model_and_assets(c_m=c_m, p_m=p_m, p_l_1=p_l, p_r=p_r, dt=dt))
task = Balance(swing_up=True, sparse=False, random=random)
environment_kwargs = environment_kwargs or {}
return control.Environment(
physics, task, time_limit=time_limit, **environment_kwargs)
def balance(time_limit=_DEFAULT_TIME_LIMIT, random=None,
environment_kwargs=None):
"""Returns the Cartpole Balance task."""
physics = Physics.from_xml_string(*get_model_and_assets())
task = Balance(swing_up=False, sparse=False, random=random)
environment_kwargs = environment_kwargs or {}
return control.Environment(
physics, task, time_limit=time_limit, **environment_kwargs)
def easy(time_limit=_DEFAULT_TIME_LIMIT, random=None, environment_kwargs=None, **kwargs):
"""Returns the easy cloth task."""
physics = Physics.from_xml_string(*get_model_and_assets())
task = Rope(randomize_gains=False, random=random, **kwargs)
environment_kwargs = environment_kwargs or {}
return control.Environment(
physics, task, time_limit=time_limit,n_frame_skip=1, rope_task=True,**environment_kwargs)
def stand(time_limit=_DEFAULT_TIME_LIMIT, random=None, environment_kwargs=None):
"""Returns a Hopper that strives to stand upright, balancing its pose."""
physics = Physics.from_xml_string(*get_model_and_assets())
task = Hopper(hopping=False, random=random)
environment_kwargs = environment_kwargs or {}
return control.Environment(
physics, task, time_limit=time_limit, control_timestep=_CONTROL_TIMESTEP,
**environment_kwargs)
def walk(time_limit=_DEFAULT_TIME_LIMIT, random=None):
"""Returns the Walk task."""
physics = Physics.from_xml_string(*get_model_and_assets())
task = Humanoid(move_speed=_WALK_SPEED, pure_state=False, random=random)
return control.Environment(physics,
task,
time_limit=time_limit,
control_timestep=_CONTROL_TIMESTEP)
def stand(time_limit=_DEFAULT_TIME_LIMIT, random=None, environment_kwargs=None):
"""Returns the Stand task."""
physics = Physics.from_xml_string(*get_model_and_assets())
task = HumanoidCMU(move_speed=0, random=random)
environment_kwargs = environment_kwargs or {}
return control.Environment(
physics, task, time_limit=time_limit, control_timestep=_CONTROL_TIMESTEP,
**environment_kwargs)
def swingup(time_limit=_DEFAULT_TIME_LIMIT, random=None,
environment_kwargs=None, setting_kwargs=None):
"""Returns the Cartpole Swing-Up task."""
physics = Physics.from_xml_string(*common.settings.get_model_and_assets_from_setting_kwargs('cartpole.xml', setting_kwargs))
task = Balance(swing_up=True, sparse=False, random=random)
environment_kwargs = environment_kwargs or {}
return control.Environment(
physics, task, time_limit=time_limit, **environment_kwargs)
def stand(time_limit=_DEFAULT_TIME_LIMIT, random=None):
"""Returns the Stand task."""
physics = Physics.from_xml_string(*get_model_and_assets())
task = PlanarWalker(move_speed=0, random=random)
return control.Environment(physics,
task,
time_limit=time_limit,
control_timestep=_CONTROL_TIMESTEP)
def turn_hard(time_limit=_DEFAULT_TIME_LIMIT, random=None):
"""Returns the hard Turn task."""
physics = Physics.from_xml_string(*get_model_and_assets())
task = Turn(target_radius=_HARD_TARGET_SIZE, random=random)
return control.Environment(physics,
task,
time_limit=time_limit,
control_timestep=_CONTROL_TIMESTEP)
def swim(time_limit=_DEFAULT_TIME_LIMIT, random=None):
"""Returns the Fish Swim task."""
physics = Physics.from_xml_string(*get_model_and_assets())
task = Swim(random=random)
return control.Environment(physics,
task,
control_timestep=_CONTROL_TIMESTEP,
time_limit=time_limit)
def run(time_limit=_DEFAULT_TIME_LIMIT, random=None, environment_kwargs=None):
"""Returns the Run task."""
physics = Physics.from_xml_string(*get_model_and_assets())
task = PlanarWalker(move_speed=_RUN_SPEED, random=random)
environment_kwargs = environment_kwargs or {}
return control.Environment(
physics, task, time_limit=time_limit, control_timestep=_CONTROL_TIMESTEP,
**environment_kwargs)
def swingup_sparse(time_limit=_DEFAULT_TIME_LIMIT, random=None,
environment_kwargs=None):
"""Returns Acrobot sparse balance."""
physics = Physics.from_xml_string(*get_model_and_assets())
task = Balance(sparse=True, random=random)
environment_kwargs = environment_kwargs or {}
return control.Environment(
physics, task, time_limit=time_limit, **environment_kwargs)
def hard(
time_limit=_DEFAULT_TIME_LIMIT, dynamics_seed=None, visual_seed=None, vary=DMCR_VARY
):
model = get_model(visual_seed, vary)
assets, _ = get_assets(visual_seed, vary)
physics = Physics.from_xml_string(model, assets)
task = Reacher(target_size=_SMALL_TARGET, random=dynamics_seed)
return control.Environment(physics, task, time_limit=time_limit,)
def hop(time_limit=_DEFAULT_TIME_LIMIT, random=None, environment_kwargs=None):
"""Returns a Hopper that strives to hop forward."""
physics = Physics.from_xml_string(*get_model_and_assets())
task = Hopper(hopping=True, random=random)
environment_kwargs = environment_kwargs or {}
return control.Environment(
physics, task, time_limit=time_limit, control_timestep=_CONTROL_TIMESTEP,
**environment_kwargs)
def swingup(
time_limit=_DEFAULT_TIME_LIMIT, dynamics_seed=None, visual_seed=None, vary=DMCR_VARY
):
model = get_model(visual_seed, vary)
assets, _ = get_assets(visual_seed, vary)
physics = Physics.from_xml_string(model, assets)
task = Balance(swing_up=True, sparse=False, random=dynamics_seed)
return control.Environment(physics, task, time_limit=time_limit)
def run(time_limit=_DEFAULT_TIME_LIMIT, random=None):
"""Returns the Run task."""
physics = Physics.from_xml_string(*get_model_and_assets())
task = HumanoidCMU(move_speed=_RUN_SPEED, random=random)
return control.Environment(physics,
task,
time_limit=time_limit,
control_timestep=_CONTROL_TIMESTEP)
def collide(time_limit=_DEFAULT_TIME_LIMIT, random=None, environment_kwargs=None):
"""Returns the Collide task."""
physics = Physics.from_xml_string(*get_model_and_assets())
task = Simple()
environment_kwargs = environment_kwargs or {}
return control.Environment(
physics, task, time_limit=time_limit, control_timestep=_CONTROL_TIMESTEP,
**environment_kwargs)
def swingup(time_limit=_DEFAULT_TIME_LIMIT, random=None,
environment_kwargs=None):
"""Returns pendulum swingup task ."""
physics = Physics.from_xml_string(*get_model_and_assets())
task = SwingUp(random=random)
environment_kwargs = environment_kwargs or {}
return control.Environment(
physics, task, time_limit=time_limit, **environment_kwargs)
def catch(time_limit=_DEFAULT_TIME_LIMIT, random=None):
"""Returns the Ball-in-Cup task."""
physics = Physics.from_xml_string(*get_model_and_assets())
task = BallInCup(random=random)
return control.Environment(physics,
task,
time_limit=time_limit,
control_timestep=_CONTROL_TIMESTEP)
def mocap(time_limit=_DEFAULT_TIME_LIMIT, random=None, environment_kwargs=None):
"""Returns the mocap task."""
physics = mujoco.Physics.from_xml_string(*get_model_and_assets())
task = jeffRat(random=random)
environment_kwargs = environment_kwargs or {}
return control.Environment(
physics, task, time_limit=time_limit, n_sub_steps=_CONTROL_SUBSTEPS,
**environment_kwargs)
def chase(time_limit=_TIME_LIMIT, random=None, environment_kwargs=None):
"""Returns manipulator bring task with the ball prop."""
physics = Physics.from_xml_string(*make_model(False, False))
task = Chase(use_peg=False, insert=False, fully_observable=True, random=random)
environment_kwargs = environment_kwargs or {}
return control.Environment(
physics, task, control_timestep=_CONTROL_TIMESTEP, time_limit=time_limit,
**environment_kwargs)
def three_poles(time_limit=_DEFAULT_TIME_LIMIT, random=None, num_poles=3,
sparse=False, environment_kwargs=None):
"""Returns the Cartpole Balance task with three or more poles."""
physics = Physics.from_xml_string(*get_model_and_assets(num_poles=num_poles))
task = Balance(swing_up=True, sparse=sparse, random=random)
environment_kwargs = environment_kwargs or {}
return control.Environment(
physics, task, time_limit=time_limit, **environment_kwargs)
def swingup_acrobot(time_limit=_DEFAULT_TIME_LIMIT, random=None,
environment_kwargs=None, p_m=1., p_l_1=.5, p_l_2=None, p_r=.045, dt=.1):
"""Returns Acrobot balance task."""
physics = Physics.from_xml_string(*get_model_and_assets(p_m=p_m, p_l_1=p_l_1, p_l_2=p_l_2, p_r=p_r, dt=dt))
task = Balance(sparse=False, random=random)
environment_kwargs = environment_kwargs or {}
return control.Environment(
physics, task, time_limit=time_limit, **environment_kwargs)
def swingup_sparse(time_limit=_DEFAULT_TIME_LIMIT, random=None,
environment_kwargs=None):
"""Returns the sparse reward variant of teh Cartpole Swing-Up task."""
physics = Physics.from_xml_string(*get_model_and_assets())
task = Balance(swing_up=True, sparse=True, random=random)
environment_kwargs = environment_kwargs or {}
return control.Environment(
physics, task, time_limit=time_limit, **environment_kwargs)
def fetch(time_limit=_DEFAULT_TIME_LIMIT, random=None, environment_kwargs=None):
"""Returns the Fetch task."""
physics = Physics.from_xml_string(*get_model_and_assets(remove_ball=False))
task = Fetch(random=random)
environment_kwargs = environment_kwargs or {}
return control.Environment(physics, task, time_limit=time_limit,
control_timestep=_CONTROL_TIMESTEP,
**environment_kwargs)
