-
Notifications
You must be signed in to change notification settings - Fork 10
/
teleop_collect_data.py
169 lines (138 loc) · 6.95 KB
/
teleop_collect_data.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
from pathlib import Path
import numpy as np
import sapien.core as sapien
import transforms3d.euler
from hand_detector.hand_monitor import Record3DSingleHandMotionControl
from hand_teleop.env.sim_env.constructor import add_default_scene_light
from hand_teleop.env.sim_env.mug_flip_env import MugFlipEnv
from hand_teleop.env.sim_env.relocate_env import RelocateEnv
from hand_teleop.env.sim_env.table_door_env import TableDoorEnv
from hand_teleop.gui.teleop_gui import GUIBase, DEFAULT_TABLE_TOP_CAMERAS
from hand_teleop.kinematics.mano_robot_hand import MANORobotHand
from hand_teleop.player.recorder import DataRecorder
def main():
# Choose a task: relocate, open_door, flip
task_name = ["open_door", "relocate", "flip"][0]
# Choose object if you are working relocate
object_name = ['tomato_soup_can', 'bleach_cleanser', 'mug', "mustard_bottle", "potted_meat_can"][3]
# Setup
frame_skip = 5
object_scale = 0.8
if task_name == "relocate":
task_full_name = f"relocate-{object_name}"
env_dict = dict(frame_skip=frame_skip, object_name=object_name, object_scale=object_scale)
elif task_name == "open_door":
task_full_name = "table_door"
env_dict = dict(frame_skip=frame_skip)
elif task_name == "flip":
task_full_name = "flip"
env_dict = dict(frame_skip=frame_skip)
else:
raise NotImplementedError
# Specify the demonstration file path and name
demo_data_root_path = Path(__file__).parent / "teleop_collected_data"
demo_data_root_path.mkdir(exist_ok=True)
demo_index = "0000" # demo_index, used only in the name of the demonstration file
path = Path(demo_data_root_path) / task_full_name
path = path / f"{demo_index}.pickle"
if task_name == "open_door":
env = TableDoorEnv(**env_dict, use_gui=True)
elif task_name == "relocate":
env = RelocateEnv(**env_dict, use_gui=True)
elif task_name == "flip":
env = MugFlipEnv(**env_dict, use_gui=True)
else:
raise NotImplementedError
env.reset_env()
env.seed(int(demo_index))
# Setup viewer and camera
add_default_scene_light(env.scene, env.renderer)
gui = GUIBase(env.scene, env.renderer)
for name, params in DEFAULT_TABLE_TOP_CAMERAS.items():
gui.create_camera(**params)
gui.viewer.set_camera_rpy(0, -0.7, 0.01)
gui.viewer.set_camera_xyz(-0.4, 0, 0.45)
scene = env.scene
viz_mat_hand_init = gui.context.create_material(np.array([0, 0, 0, 0]), np.array([0.96, 0.75, 0.69, 1]), 0.0, 0.8,
0)
# Perception
motion_control = Record3DSingleHandMotionControl(hand_mode="right_hand", show_hand=True)
# Recorder
recorder = DataRecorder(filename=str(path.resolve()), scene=scene)
# Init
create_robot = False
steps = 0
env_init_pos = np.array([-0.4, 0, 0.2])
rgb, depth = motion_control.camera.fetch_rgb_and_depth()
locked_indices = []
scene.step()
# Press "q" on the keyboard to exit the teleoperation when you finish
# The demonstration data will be automatically saved
while not gui.closed:
for _ in range(frame_skip):
scene.step()
gui.render(additional_views=[rgb[..., ::-1]])
steps += 1
if not motion_control.initialized:
success, motion_data = motion_control.step()
rgb = motion_data["rgb"]
if not success:
continue
viz_mat_hand_init.set_base_color(motion_control.init_process_color)
rotate_pose = sapien.Pose(q=[0.9238, 0, 0.3826, 0], p=[0.2, 0, -0.1])
gui.update_mesh(motion_data["vertices"], motion_data["faces"], viz_mat=viz_mat_hand_init,
clear_context=True, pose=sapien.Pose(env_init_pos) * rotate_pose)
else:
if not create_robot:
zero_joint_pos = motion_control.compute_hand_zero_pos()
mano_robot = MANORobotHand(env.scene, env.renderer, init_joint_pos=zero_joint_pos,
control_interval=frame_skip * scene.get_timestep(), scale=1)
robot = mano_robot.robot
robot.set_pose(sapien.Pose(env_init_pos, transforms3d.euler.euler2quat(0, np.pi / 2, 0)))
create_robot = True
# Lock means that the finger will not move regardless of the hand pose detection results
# It can save you sometime when you already grasp something and do not want to release it
# You can press "z" on the keyboard to lock the hand and then press "x" to unlock it
def change_locked():
locked_indices.clear()
contact_finger_indices = mano_robot.check_contact_finger([env.target_object])
locked_indices.extend(contact_finger_indices)
mano_robot.highlight_finger_color(contact_finger_indices)
def clear_locked():
locked_indices.clear()
mano_robot.clear_finger_color()
gui.register_keydown_action('z', change_locked)
gui.register_keydown_action('x', clear_locked)
# Clear colored hand visualization during initialization
for i in range(len(gui.nodes)):
node = gui.nodes.pop()
gui.render_scene.remove_node(node)
success, motion_data = motion_control.step()
rgb = motion_data["rgb"]
# Data recording.py
record_data = motion_data.copy()
record_data.update({"success": success})
# Remove the pop code if you want to save the camera image in into the dataset. It can be large.
record_data.pop("rgb")
record_data.pop("depth")
recorder.step(record_data)
if not success:
continue
root_joint_qpos = motion_control.compute_operator_space_root_qpos(motion_data)
root_joint_qpos *= 1
finger_joint_qpos = mano_robot.compute_qpos(motion_data["pose_params"][3:])
robot_qpos = np.concatenate([root_joint_qpos, finger_joint_qpos])
if np.abs(robot.get_qpos().mean()) < 1e-5:
robot.set_qpos(robot_qpos)
mano_robot.control_robot(robot_qpos, confidence=motion_data["confidence"], lock_indices=locked_indices)
# Create SAPIEN mesh for rendering
# gui.update_mesh(motion_data["vertices"], motion_data["faces"], viz_mat=viz_mat_hand_init,
# clear_context=True,
# pose=sapien.Pose(root_joint_qpos[:3] + np.array([0, -0.5, 0]) + env_init_pos))
print(len(recorder.data_list))
meta_data = dict(env_class=env.__class__.__name__, env_kwargs=env_dict,
shape_param=motion_control.calibrated_shape_params,
zero_joint_pos=motion_control.compute_hand_zero_pos())
recorder.dump(meta_data)
if __name__ == '__main__':
main()