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/mujoco/%s.npy" % env_name
horizon = ENV_PARAMS[env_name]['max_path_length']
collect_demos_fixed(
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 gym
import numpy as np
if __name__ == '__main__':
expert = SpaceMouseExpert(
xyz_dims=1,
xyz_remap=[1, 0, 2],
xyz_scale=[1, -1, -1],
)
env = gym.make("MountainCarContinuous-v0")
env.render()
# env = SawyerMultiobjectEnv(
# num_objects=1,
# preload_obj_dict=[
# dict(color2=(0.1, 0.1, 0.9)),
# ],
# )
# env = ImageEnv(env,
# recompute_reward=False,
# # transpose=True,
# init_camera=sawyer_pusher_camera_upright_v2,
# )
collect_demos_fixed(env,
expert,
"test.npy",
100,
horizon=10000,
render=True)
import gym
import numpy as np
from sawyer_control.envs.sawyer_insertion_refined import SawyerHumanControlEnv
if __name__ == '__main__':
expert = KeyboardExpert(
xyz_dims=3,
# xyz_remap=[0, 1, 2],
xyz_scale=[-1, -1, 0.75],
)
env = SawyerHumanControlEnv(action_mode='joint_space_impd',
position_action_scale=1,
max_speed=0.015)
# env = gym.make("MountainCarContinuous-v0")
# env = SawyerMultiobjectEnv(
# num_objects=1,
# preload_obj_dict=[
# dict(color2=(0.1, 0.1, 0.9)),
# ],
# )
# env = ImageEnv(env,
# recompute_reward=False,
# # transpose=True,
# init_camera=sawyer_pusher_camera_upright_v2,
# )
collect_demos_fixed(env, expert, "insertion10.npy", 10)
# data = load_local_or_remote_file(
# '/home/murtaza/research/railrl/data/doodads3/03-08-bc-ant-gym-v1/03-08-bc_ant_gym_v1_2020_03_08_19_22_00_id000--s39483/bc.pkl')
# # env = gym.make('Ant-v2')
# policy = MakeDeterministic(data.cpu())
# collect_demos_fixed(env, policy, "data/local/demos/ant_off_policy_10_demos_100.npy", N=100, horizon=1000, threshold=-1,
# render=False)
data = load_local_or_remote_file(
'/home/murtaza/research/railrl/data/local/03-09-bc-ant-frac-trajs-sweep/03-09-bc_ant_frac_trajs_sweep_2020_03_09_17_58_01_id000--s71624/bc.pkl'
)
env = gym.make('Ant-v2')
policy = data.cpu()
collect_demos_fixed(env,
policy,
"data/local/demos/ant_off_policy_10_demos_100.npy",
N=100,
horizon=1000,
threshold=-1,
render=False)
data = load_local_or_remote_file(
'/home/murtaza/research/railrl/data/local/03-09-bc-ant-frac-trajs-sweep/03-09-bc_ant_frac_trajs_sweep_2020_03_09_17_58_02_id000--s47768/bc.pkl'
)
env = gym.make('Ant-v2')
policy = data.cpu()
collect_demos_fixed(env,
policy,
"data/local/demos/ant_off_policy_15_demos_100.npy",
N=100,
horizon=1000,
threshold=-1,
# 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 SawyerMultiobjectEnv
# from multiworld.envs.pygame.point2d import Point2DWallEnv
import gym
import numpy as np
if __name__ == '__main__':
expert = SpaceMouseExpert(
xyz_dims=1,
xyz_remap=[1, 0, 2],
xyz_scale=[-1, -1, -1],
)
env = gym.make("MountainCarContinuous-v0")
# env = SawyerMultiobjectEnv(
# num_objects=1,
# preload_obj_dict=[
# dict(color2=(0.1, 0.1, 0.9)),
# ],
# )
# env = ImageEnv(env,
# recompute_reward=False,
# # transpose=True,
# init_camera=sawyer_pusher_camera_upright_v2,
# )
collect_demos_fixed(env, expert, "test.npy", 10)
env = ImageEnv(
env,
recompute_reward=False,
transpose=True,
image_length=450000,
reward_type="image_distance",
# init_camera=sawyer_pusher_camera_upright_v2,
)
# env.reset()
for i in range(25):
collect_demos_fixed(
env,
expert,
"/home/anair/ros_ws/src/railrl-private/demos/demo_v4_grey_%i.pkl" %
i,
1,
horizon=1000,
pause=0.05)
print("Collected demo: ", i)
# for i in range(10):
# collect_demos_fixed(env, expert, "demos/demo_v3.pkl", 1, horizon=1000, pause=0.05)
# o = None
# while True:
# a, valid, reset, accept = expert.get_action(o)
# if valid:
# o, r, done, info = env.step(a)
# time.sleep(0.05)
# data = load_local_or_remote_file('01-12-sac-mujoco-envs/01-12-sac_mujoco_envs_2020_01_12_22_34_03_id000--s52204/params.pkl')
# data = load_local_or_remote_file('02-20-sac-mujoco-envs-unnormalized-run-longer/02-20-sac_mujoco_envs_unnormalized_run_longer_2020_02_20_23_55_13_id000--s39214/params.pkl')
# # env = data['exploration/env']
# env = gym.make('HalfCheetah-v2')
# policy = data['exploration/policy']
# collect_demos_fixed(env, policy, "data/local/demos/hc_action_noise_1000.npy", N=1000, horizon=1000, threshold=9000, render=False)
#
data = load_local_or_remote_file(
'02-17-sac-mujoco-envs-unnormalized/02-17-sac_mujoco_envs_unnormalized_2020_02_18_01_07_08_id005--s24204/params.pkl'
)
env = data['evaluation/env']
policy = data['exploration/policy']
collect_demos_fixed(env,
policy,
"data/local/demos/hopper_action_noise_1000.npy",
N=1000,
horizon=1000,
threshold=3000,
render=False)
data = load_local_or_remote_file(
'02-17-sac-mujoco-envs-unnormalized/02-17-sac_mujoco_envs_unnormalized_2020_02_18_01_07_32_id003--s29410/params.pkl'
)
env = data['evaluation/env']
policy = data['exploration/policy']
collect_demos_fixed(env,
policy,
"data/local/demos/ant_action_noise_1000.npy",
N=1000,
horizon=1000,
threshold=5000,
from railrl.demos.collect_demo import collect_demos_fixed
from railrl.misc.asset_loader import load_local_or_remote_file
import gym
if __name__ == '__main__':
data = load_local_or_remote_file(
'02-20-sac-mujoco-envs-unnormalized-run-longer/02-20-sac_mujoco_envs_unnormalized_run_longer_2020_02_20_23_55_13_id000--s39214/params.pkl'
)
env = data['exploration/env']
policy = data['exploration/policy']
collect_demos_fixed(env,
policy,
"data/local/demos/hc_action_noise_25.npy",
N=25,
horizon=1000,
threshold=9000,
render=False)
# data = load_local_or_remote_file(
# '/home/murtaza/research/railrl/data/local/03-04-bc-hc-v2/03-04-bc_hc_v2_2020_03_04_17_57_54_id000--s90897/bc.pkl')
# env = gym.make('HalfCheetah-v2')
# policy = data.cpu()
# collect_demos_fixed(env, policy, "data/local/demos/hc_off_policy_100.npy", N=100, horizon=1000, threshold=8000,
# render=False)
# data = load_local_or_remote_file(
# '/home/murtaza/research/railrl/data/doodads3/03-05-bc-hc-gym-v5/03-05-bc_hc_gym_v5_2020_03_06_06_55_43_id000--s42378/bc.pkl')
# env = gym.make('HalfCheetah-v2')
# policy = data.cpu()
# collect_demos_fixed(env, policy, "data/local/demos/hc_off_policy_10_demos_100.npy", N=100, horizon=1000, threshold=-1,
# render=False)
# data = load_local_or_remote_file(
