def __init__(self, *args, **kwargs):
""" Constants:
omega is always 0, and set a constant for background noise
"""
self.Om = np.array([[1,0,0],[0,1,0],[0,0,1]])
self.background = 2.0 #background noise
""" Variables:
These variables will be read along with the action:
two_theta: detector's rotation about the z-axis -- assume elastic scattering, so omega is always 0
theta: the angle at which our neutrons strike the plane
These variables are the two dimensions of our problem
chi: outer ring's rotation about the x-axis
phi: rotation of the eulerian cradle, varies between z- and y-axis rotation depending on how much chi rotated
"""
self.max_two_theta = 180
self.max_chi = 90
self.max_phi = 360
self.min_chi = -90
self.min_phi = 0
self.hit = 0
#Set up hkl and all actions
super(UBEnv, self).__init__(self.model_path("UB.xml.mako"),*args, **kwargs)
#Two independent bodies
self.ring = find_body(self.world, "ring") #chi
self.eu_cradle = find_body(self.world, "eu_cradle") #phi
self.detector = find_body(self.world, "detector") #theta
self.pivot = find_joint(self.world, "angular_axis") #pivot that enables angular movement
Serializable.__init__(self, *args, **kwargs)
def __init__(self, *args, log_scale_limit=2.0, fix_params=False, rand_params=RAND_PARAMS, random_seed=None, fixed_goal=True, max_path_length=0, **kwargs):
"""
Half-Cheetah environment with randomized mujoco parameters
:param log_scale_limit: lower / upper limit for uniform sampling in logspace of base 2
:param random_seed: random seed for sampling the mujoco model params
:param fix_params: boolean indicating whether the mujoco parameters shall be fixed
:param rand_params: mujoco model parameters to sample
"""
assert set(rand_params) <= set(self.RAND_PARAMS_EXTENDED), \
"rand_params must be a subset of " + str(self.RAND_PARAMS_EXTENDED)
self.log_scale_limit = log_scale_limit
self.random_seed = random_seed
self.random_state = np.random.RandomState(random_seed)
self.fix_params = fix_params # can be changed by calling the fix_mujoco_parameters method
self.rand_params = rand_params
self.fixed_goal = fixed_goal
self.parameters_already_fixed = False
self.n_steps = 0
self.reward_range = None
self.metadata = {}
if max_path_length is not None:
self.max_path_length = max_path_length
else:
self.max_path_length = 10**8 #set to a large number
args_all, kwargs_all = get_all_function_arguments(self.__init__, locals())
Serializable.__init__(*args_all, **kwargs_all)
def __init__(self, goal_vel=None, *args, **kwargs):
self.goal_vel = goal_vel
super(HalfCheetahEnvRandDirec, self).__init__(*args, **kwargs)
self.goal_vel = goal_vel
Serializable.__init__(self, *args, **kwargs)
self.goal_vel = goal_vel
self.reset(reset_args=goal_vel)
def __init__(self,
xml_file=None,
distractors=True,
onehot=True,
*args,
**kwargs):
utils.EzPickle.__init__(self)
print("using xml_file", xml_file)
if xml_file is None:
xml_file = 'pusher.xml'
self.__class__.FILE = xml_file
self.include_distractors = distractors
self.test_dir = "/home/kevin/rllab/pushing/test2_paired_push_demos_noimg/"
self.train_dir = "/home/kevin/FaReLI_data/pushing/paired_push_demos_noimg/"
self.xml_dir = "/home/kevin/gym/gym/envs/mujoco/assets/sim_push_xmls/"
self.goal_num = 1
self.test = False
self.onehot = onehot
self.onehot_dim = 5
self.onehot_position = 0
super(PusherEnv, self).__init__(*args, **kwargs)
# import pdb; pdb.set_trace()
# self = mujoco_env.MujocoEnv(file_path=xml_file)
# self.viewer_setup()
# self.reset()
# self.observation_space = self.observation_space
# self.action_space = self.action_space
# self.goal_num, self.test = self.sample_goals(num_goals=1, test=False)[0]
Serializable.__init__(self, *args, **kwargs)
def __init__(self,
*args,
log_scale_limit=2.0,
fix_params=False,
fixed_goal=True,
random_seed=None,
**kwargs):
self.sign = 1
self.first = True
self.penalty_ctrl = 0.01
self.init_geom_size = None
self.init_body_pos = None
self.init_geom_pos = None
args_all, kwargs_all = get_all_function_arguments(
self.__init__, locals())
BaseEnvRandParams.__init__(self,
log_scale_limit=log_scale_limit,
fix_params=fix_params,
rand_params=['geom_size'],
fixed_goal=fixed_goal,
random_seed=random_seed)
MujocoEnv.__init__(self, *args, **kwargs)
Serializable.__init__(*args_all, **kwargs_all)
def __init__(self,
box_center=(0, 0),
box_noise=0,
frame_skip=5,
scale=0.05,
reward_delay=1,
timestep=0.01,
*args,
**kwargs):
FILE = osp.join(osp.abspath(osp.dirname(__file__)),
'assets/throw_pos.xml')
self.box_center = box_center
self.box_noise = box_noise
self.relativeBoxPosition = np.zeros(3)
self.init_block_goal_dist = 1.0
self.scale = scale
self.reward_delay = reward_delay
self.count = 0
self.low = np.array(
[-0.65, -1.7, -1.4, -1.8, -2.6, -1.8, -0.26, -0.35, -0.24])
self.high = np.array([0.68, 1.5, 1, 1.8, 1.5, 1.8, 1.2, 1, 1])
super(ThrowerPosEnv, self).__init__(model_path=FILE,
frame_skip=frame_skip,
*args,
**kwargs)
self.sim.model.opt.timestep = timestep
Serializable.__init__(self, box_center, box_noise, frame_skip, *args,
**kwargs)
self.init_qpos[:9] = np.array(
[-0.07855885, -1.5, 0.6731897, 1.5, -2.83926611, 0, 0.9, 0.9, 0.9])
self.action_space = spaces.Box(low=-np.ones(9) * 5,
high=np.ones(9) * 5,
dtype=np.float32)
def __init__(self, box_center=(0,0), box_noise=0.4, frame_skip=5, *args, **kwargs):
self.box_center = box_center
self.box_noise = box_noise
super(LobberEnv, self).__init__(frame_skip=frame_skip, *args, **kwargs)
Serializable.__init__(self, box_center, box_noise, frame_skip, *args, **kwargs)
def __init__(self,
xml_file=None,
distance_metric_order=None,
distractors=True,
*args,
**kwargs):
self.goal = None
if 'noise' in kwargs:
noise = kwargs['noise']
else:
noise = 0.0
self.include_distractors = distractors
if self.include_distractors:
self.shuffle_order = [[0, 1, 2], [1, 2, 0], [2, 0, 1]][2]
# self.shuffle_order = rd.sample([[0,1,2],[1,2,0],[2,0,1]],1)[0]
if xml_file is None:
if not self.include_distractors:
xml_file = '/home/kevin/maml_rl/vendor/mujoco_models/r7dof_versions/reacher_7dof.xml'
else:
xml_file = '/home/kevin/maml_rl/vendor/mujoco_models/r7dof_versions/reacher_7dof_2distr_%s%s%s.xml' % tuple(
self.shuffle_order)
print("xml file", xml_file)
self.mujoco = mujoco_env.MujocoEnv(file_path=xml_file,
action_noise=noise)
self.action_space = self.mujoco.action_space
self.get_viewer = self.mujoco.get_viewer
self.log_diagnostics = self.mujoco.log_diagnostics
Serializable.__init__(self, *args, **kwargs)
def __init__(self,
goal_args=('noisy', (.6, .2), .1),
frame_skip=5,
scale=0.05,
reward_delay=1,
timestep=0.01,
*args,
**kwargs):
FILE = osp.join(osp.abspath(osp.dirname(__file__)),
'assets/picker_pos.xml')
self.goal_args = goal_args
self.numClose = 0
self.reward_delay = reward_delay
self.count = 0
self.scale = scale
self.low = np.array(
[-0.65, -2, -1.4, -1.55, -2.6, -1.8, -0.26, -0.35, -0.24])
self.high = np.array([0.68, 0.86, 1, 1.75, 1.5, 1.8, 1.5, 1.5, 1.5])
super(PickerPosEnv, self).__init__(model_path=FILE,
frame_skip=frame_skip,
*args,
**kwargs)
self.init_qpos[:9] = np.array(
[-0.07855885, -1.5, 0.6731897, 1.5, -2.83926611, 0, 0.2, 0.2, 0.2])
self.sim.model.opt.timestep = timestep
Serializable.__init__(self, goal_args, frame_skip, *args, **kwargs)
def __init__(self, *args, **kwargs):
self.n_arms = kwargs.get('n_arms')
self.max_path_length = kwargs.get('max_path_length')
self.arm_means = np.zeros((self.n_arms))
self.ts = 0
super(MABEnv, self).__init__(*args, **kwargs)
Serializable.__init__(self, *args, **kwargs)
def __init__(self, distance_metric_order=None, *args, **kwargs):
self.goal = None
self.shuffle_order = None
self.objects = ["goal", "distract1", "distract2"]
self.__class__.FILE = 'r7dof_versions/reacher_7dof_2distract.xml'
seed = kwargs['envseed']
super().__init__(envseed=seed)
Serializable.__init__(self, *args, **kwargs)
def __init__(self, *args, **kwargs):
super(PegEnv, self).__init__(*args, **kwargs)
Serializable.__init__(self, *args, **kwargs)
self.PR2_GAINS = np.array(
[3.09, 1.08, 0.393, 0.674, 0.111, 0.152, 0.098])
self.t = 0
self.T = 100
self.init_body_pos = self.model.body_pos.copy()
def __init__(self,
task={},
n_tasks=100,
randomize_tasks=True,
goal_radius=0.8,
**kwargs):
self.goal_radius = goal_radius
Serializable.__init__(self)
super(AntGoalEnvSparse, self).__init__(task, n_tasks, **kwargs)
def __init__(self, distance_metric_order=None, *args, **kwargs):
self.goal = None
self.__class__.FILE = 'r7dof_versions/reacher_7dof_sparse0.2.xml'
if "envseed" in kwargs.keys():
seed = kwargs['envseed']
else:
seed = 0
super().__init__(envseed=seed)
Serializable.__init__(self, *args, **kwargs)
def __init__(
self, start_pos=(.1,1.7), start_noise=0.2, frame_skip=2,
*args, **kwargs):
self.start_pos = start_pos
self.start_noise = start_noise
super(CatchEnv, self).__init__(frame_skip=frame_skip, *args, **kwargs)
Serializable.__init__(self, start_pos, start_noise, frame_skip, *args, **kwargs)
def __init__(self,
goal_args=('noisy', (.6, .2), .1),
frame_skip=5,
*args,
**kwargs):
self.goal_args = goal_args
super(PickerEnv, self).__init__(frame_skip=frame_skip, *args, **kwargs)
Serializable.__init__(self, goal_args, frame_skip, *args, **kwargs)
def __init__(self, *args, **kwargs):
super(CartpoleSwingupEnvX, self).__init__(
self.model_path("cartpole.xml.mako"),
*args, **kwargs
)
self.max_cart_pos = 3
self.max_reward_cart_pos = 3
self.cart = find_body(self.world, "cart")
self.pole = find_body(self.world, "pole")
Serializable.__init__(self, *args, **kwargs)
def __init__(self, option=default_reacher_env_option, *args, **kwargs):
self.goal = None
if 'noise' in kwargs:
noise = kwargs['noise']
else:
noise = 0.0
print("using env option", ENV_OPTIONS[option])
self.__class__.FILE = ENV_OPTIONS[option]
super().__init__(action_noise=noise) # file_path=ENV_OPTIONS[option])
Serializable.__init__(self, *args, **kwargs)
def __init__(self, num_goals=40, train=True, *args, **kwargs):
thetas = np.linspace(0, np.pi / 2, 40)
self.goals = np.array([[2 * np.cos(theta), 2 * np.sin(theta)]
for theta in thetas])
self.goal = None
self.num_goals = num_goals
self.sparse = False
self.info_logKeys = ['goal_dist']
super(continious_test, self).__init__(*args, **kwargs)
Serializable.__init__(self, *args, **kwargs)
def __init__(self, goal=None, mode='local', *args, **kwargs):
self._goal_idx = None
#if mode == 'local':
#self.goals = pickle.load(open("/home/russellm/iclr18/maml_rosen/rllab/envs/mujoco/ant_10Deg.pkl", "rb"))
#self.goals = joblib.load('/home/russellm/iclr18/rosen_data/saved_expert_traj/Expert_trajs_dense_ant/goals_pool.pkl')['goals_pool']
self.goals = pickle.load(
open("/home/russellm/mri_onPolicy/rllab/envs/mujoco/goals_ant.pkl",
'rb'))
super(AntEnvRandGoalRing, self).__init__(*args, **kwargs)
Serializable.__init__(self, *args, **kwargs)
def __init__(self, *args, **kwargs):
#self.dt = .1 # refresh rate
#self.on_goal = 0
super(ArmEnv, self).__init__(self.model_path("arm.xml.mako"), *args,
**kwargs)
self.arm1 = find_body(self.world, "arm1")
self.arm2 = find_body(self.world, "arm2")
self.point = find_body(self.world, "point")
Serializable.__init__(self, *args, **kwargs)
def __init__(self, xml_file, *args, **kwargs):
self.frame_skip = 5
self.__class__.FILE = xml_file
if 'distractors' in kwargs:
self.include_distractors = kwargs['distractors']
else:
self.include_distractors = False
#kwargs.pop('xml_file')
super(PusherEnv, self).__init__(*args, **kwargs)
self.frame_skip = 5
Serializable.__init__(self, *args, **kwargs)
def __init__(self, goal= None, mode = 'local',*args, **kwargs):
self._goal = None
#if mode == 'local':
self.goals = pickle.load(open("/home/russellm/iclr18/maml_rosen/rllab/envs/mujoco/ant_10Deg.pkl", "rb"))
#elif mode == 'local_docker' or mode == 'ec2':
#self.goals = pickle.load(open("/root/code/rllab/rllab/envs/mujoco/goals_ant.pkl", "rb"))
# else:
# raise AssertionError('Mode needs to be local, docker or ec2')
super(AntEnvRandGoalRing, self).__init__(*args, **kwargs)
Serializable.__init__(self, *args, **kwargs)
def __init__(self, distance_metric_order=None, *args, **kwargs):
self.goal = None
self.shuffle_order = None
self.objects = ["goal", "distract1", "distract2"]
self.__class__.FILE = 'r7dof_versions/reacher_7dof_2distract.xml'
# self.__class__.FILE = '/home/rosen/FaReLI_data/pushing/push_textures/sim_push_xmls/train_ensure_woodtable_distractor_pusher25.xml'
if "envseed" in kwargs.keys():
seed = kwargs['envseed']
else:
seed = 123456
super().__init__(envseed=seed)
Serializable.__init__(self, *args, **kwargs)
def __init__(self, *args, **kwargs):
self.frame_skip = 5
if 'xml_file' in kwargs:
self.__class__.FILE = kwargs['xml_file']
if 'distractors' in kwargs:
self.include_distractors = kwargs['distractors']
else:
self.include_distractors = False
super(PusherEnv2D, self).__init__(*args, **kwargs)
self.frame_skip = 5
self.dist = []
Serializable.__init__(self, *args, **kwargs)
def __init__(self, *args, **kwargs):
self.max_pole_angle = .2
self.max_cart_pos = 2.4
self.max_cart_speed = 4.
self.max_pole_speed = 4.
super(CartpoleEnv, self).__init__(
self.model_path("cartpole.xml.mako"),
*args, **kwargs
)
self.cart = find_body(self.world, "cart")
self.pole = find_body(self.world, "pole")
Serializable.__init__(self, *args, **kwargs)
def __init__(self, *args, **kwargs):
# make sure mdp-level step is 100ms long
kwargs["frame_skip"] = kwargs.get("frame_skip", 2)
if kwargs.get("template_args", {}).get("noise", False):
self.link_len = (np.random.rand()-0.5) + 1
else:
self.link_len = 1
kwargs["template_args"] = kwargs.get("template_args", {})
kwargs["template_args"]["link_len"] = self.link_len
super(DoublePendulumEnv, self).__init__(
self.model_path("double_pendulum.xml.mako"),
*args, **kwargs
)
self.link1 = find_body(self.world, "link1")
self.link2 = find_body(self.world, "link2")
Serializable.__init__(self, *args, **kwargs)
def __init__(self, *args, **kwargs):
Serializable.__init__(self, *args, **kwargs)
self.random_start = kwargs.pop("random_start", True)
self.random_start_range = kwargs.pop("random_start_range", 1.)
super(CarParkingEnv, self).__init__(
self.model_path("car_parking.xml"),
*args, **kwargs
)
self.goal = find_body(self.world, "goal")
self.car = find_body(self.world, "car")
self.wheels = [
body for body in self.world.bodies if "wheel" in _get_name(body)]
self.front_wheels = [
body for body in self.wheels if "front" in _get_name(body)]
self.max_deg = 30.
self.goal_radius = 1.
self.vel_thres = 1e-1
self.start_radius = 5.
def __init__(self, goal= None, *args, **kwargs):
self._goal_vel = goal
#if goal is not None:
# self._goal_vel *= 100
super(AntEnvRand, self).__init__(*args, **kwargs)
Serializable.__init__(self, *args, **kwargs)
def __init__(self, *args, **kwargs):
super(AntEnvDirecOracle, self).__init__(*args, **kwargs)
Serializable.__init__(self, *args, **kwargs)
def __init__(self, goal=None, *args, **kwargs):
self._goal_vel = goal
super(HalfCheetahEnvRand, self).__init__(*args, **kwargs)
Serializable.__init__(self, *args, **kwargs)
def __init__(self, *args, **kwargs):
super(HalfCheetahEnvOracle, self).__init__(*args, **kwargs)
self._goal_vel = None
Serializable.__init__(self, *args, **kwargs)
def __init__(self, goal=None, *args, **kwargs):
self._goal_vel = goal
super(AntEnvRandDirec, self).__init__(*args, **kwargs)
Serializable.__init__(self, *args, **kwargs)
def __init__(self, goal= None, *args, **kwargs):
self._goal_pos = goal
super(AntEnvRandGoal, self).__init__(*args, **kwargs)
Serializable.__init__(self, *args, **kwargs)
def __init__(self, *args, **kwargs):
super(HalfCheetahEnvDirecOracle, self).__init__(*args, **kwargs)
Serializable.__init__(self, *args, **kwargs)
