from railrl.demos.collect_demo import collect_demos, SpaceMouseExpert
from multiworld.core.image_env import ImageEnv
from multiworld.envs.mujoco.cameras import sawyer_pusher_camera_upright_v2
from multiworld.envs.mujoco.sawyer_xyz.sawyer_push_multiobj import SawyerTwoObjectNIPSEnv
from multiworld.envs.pygame.point2d import Point2DWallEnv
import numpy as np
if __name__ == '__main__':
expert = SpaceMouseExpert(
xyz_dims=2,
xyz_remap=[1, 0, 2],
xyz_scale=[-1, -1, -1],
)
env = SawyerTwoObjectNIPSEnv()
env = ImageEnv(
env,
recompute_reward=False,
# transpose=True,
init_camera=sawyer_pusher_camera_upright_v2,
)
collect_demos(env, expert, "pusher_demos_100.npy", 100)
)
x_low = -0.2
x_high = 0.2
y_low = 0.5
y_high = 0.7
t = 0.03
env = SawyerMultiobjectEnv(
num_objects=1,
reset_to_initial_position=False,
puck_goal_low=(x_low + t + t, y_low + t),
puck_goal_high=(x_high - t - t, y_high - t),
hand_goal_low=(x_low, y_low),
hand_goal_high=(x_high, y_high),
mocap_low=(x_low, y_low, 0.0),
mocap_high=(x_high, y_high, 0.5),
object_low=(x_low + t + t, y_low + t, 0.0),
object_high=(x_high - t - t, y_high - t, 0.5),
preload_obj_dict=[
dict(color2=(0.1, 0.1, 0.9)),
],
)
env = ImageEnv(
env,
recompute_reward=False,
# transpose=True,
init_camera=sawyer_init_camera_zoomed_in,
)
collect_demos(env, expert, "pusher_reset_free_demos_100b.npy", 100)
from multiworld.envs.mujoco.cameras import sawyer_init_camera_zoomed_in
import numpy as np
from railrl.demos.collect_demo import collect_demos
from railrl.misc.asset_loader import load_local_or_remote_file
if __name__ == '__main__':
data = load_local_or_remote_file('/home/murtaza/research/railrl/data/doodads3/11-16-pusher-state-td3-sweep-params-policy-update-period/11-16-pusher_state_td3_sweep_params_policy_update_period_2019_11_17_00_28_45_id000--s62098/params.pkl')
env = data['evaluation/env']
policy = data['trainer/trained_policy']
image_env = ImageEnv(
env,
48,
init_camera=sawyer_init_camera_zoomed_in,
transpose=True,
normalize=True,
)
collect_demos(image_env, policy, "data/local/demos/pusher_demos_action_noise_1000.npy", N=1000, horizon=50, threshold=.1, add_action_noise=False, key='puck_distance', render=True, noise_sigma=0.0)
# data = load_local_or_remote_file("demos/pusher_demos_1000.npy")
# for i in range(100):
# goal = data[i]['observations'][49]['desired_goal']
# o = env.reset()
# path_length = 0
# while path_length < 50:
# env.set_goal({'state_desired_goal':goal})
# o = o['state_observation']
# new_obs = np.hstack((o, goal))
# a, agent_info = policy.get_action(new_obs)
# o, r, d, env_info = env.step(a)
# path_length += 1
# print(i, env_info['puck_distance'])
from railrl.demos.collect_demo import collect_demos, SpaceMouseExpert
from multiworld.core.image_env import ImageEnv
from multiworld.envs.mujoco.cameras import sawyer_pusher_camera_upright_v2
from multiworld.envs.pygame.point2d import Point2DWallEnv
import numpy as np
if __name__ == '__main__':
expert = SpaceMouseExpert(xyz_dims=2)
env = Point2DWallEnv(
render_onscreen=False,
images_are_rgb=True,
)
env = ImageEnv(
env,
non_presampled_goal_img_is_garbage=True,
recompute_reward=False,
# transpose=True,
# init_camera=sawyer_pusher_camera_upright_v2,
)
collect_demos(env, expert, "pointmass_demos_100.npy", 100)
policy = data['evaluation/policy']
# import ipdb; ipdb.set_trace()
# policy =
policy.to("cpu")
# image_env = ImageEnv(
# env,
# 48,
# init_camera=sawyer_init_camera_zoomed_in,
# transpose=True,
# normalize=True,
# )
# env_name = pendulum
outfile = "/home/ashvin/data/s3doodad/demos/icml2020/pusher/demos100.npy"
horizon = 200
collect_demos(
env, policy, outfile, N=100, horizon=horizon
) # , threshold=.1, add_action_noise=False, key='puck_distance', render=True, noise_sigma=0.0)
# data = load_local_or_remote_file("demos/pusher_demos_1000.npy")
# for i in range(100):
# goal = data[i]['observations'][49]['desired_goal']
# o = env.reset()
# path_length = 0
# while path_length < 50:
# env.set_goal({'state_desired_goal':goal})
# o = o['state_observation']
# new_obs = np.hstack((o, goal))
# a, agent_info = policy.get_action(new_obs)
# o, r, d, env_info = env.step(a)
# path_length += 1
# print(i, env_info['puck_distance'])
import numpy as np
from railrl.demos.collect_demo import collect_demos
from railrl.misc.asset_loader import load_local_or_remote_file
if __name__ == '__main__':
data = load_local_or_remote_file('ashvin/icml2020/murtaza/pusher/state/run3/id3/itr_980.pkl')
env = data['evaluation/env']
policy = data['trainer/trained_policy']
policy = policy.to("cpu")
image_env = ImageEnv(
env,
48,
init_camera=sawyer_init_camera_zoomed_in,
transpose=True,
normalize=True,
)
collect_demos(image_env, policy, "/home/ashvin/data/s3doodad/demos/icml2020/pusher/demos_action_noise_1000.npy", N=1000, horizon=50, threshold=.1, add_action_noise=False, key='puck_distance', render=True, noise_sigma=0.0)
# data = load_local_or_remote_file("demos/pusher_demos_1000.npy")
# for i in range(100):
# goal = data[i]['observations'][49]['desired_goal']
# o = env.reset()
# path_length = 0
# while path_length < 50:
# env.set_goal({'state_desired_goal':goal})
# o = o['state_observation']
# new_obs = np.hstack((o, goal))
# a, agent_info = policy.get_action(new_obs)
# o, r, d, env_info = env.step(a)
# path_length += 1
# print(i, env_info['puck_distance'])
if __name__ == '__main__':
data = load_local_or_remote_file(
'11-16-door-reset-free-state-td3-sweep-params-policy-update-period/11-16-door_reset_free_state_td3_sweep_params_policy_update_period_2019_11_17_00_26_50_id000--s89728/params.pkl'
)
env = data['evaluation/env']
policy = data['trainer/trained_policy']
presampled_goals_path = osp.join(
osp.dirname(mwmj.__file__),
"goals",
"door_goals.npy",
)
presampled_goals = load_local_or_remote_file(presampled_goals_path).item()
image_env = ImageEnv(
env,
48,
init_camera=sawyer_door_env_camera_v0,
transpose=True,
normalize=True,
presampled_goals=presampled_goals,
)
collect_demos(image_env,
policy,
"data/local/demos/door_demos_action_noise_1000.npy",
N=1000,
horizon=100,
threshold=.1,
add_action_noise=True,
key='angle_difference',
render=False)