-
Notifications
You must be signed in to change notification settings - Fork 0
/
gnphysics_dataset_creator.py
229 lines (202 loc) · 7.56 KB
/
gnphysics_dataset_creator.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
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
import pybullet as p
import pybullet_data
import numpy as np
import copy
import time
import pathlib
import shutil
import argparse
from typing import List, Optional
from tqdm import tqdm
from designer import *
class GNPhysicsDatasetCreator:
def __init__(
self,
designer: Designer,
dataset_prefix: str,
num_episode: int = 300,
num_frame: int = 50,
min_num_brick: int = 5,
max_num_brick: int = 25,
train_eval_split: float = 0.8,
save_dir: Optional[pathlib.Path] = None,
) -> None:
self.designer = designer
self.dataset_prefix = dataset_prefix
self.num_episode = num_episode
self.num_frame = num_frame
self.min_num_brick = min_num_brick
self.max_num_brick = max_num_brick
self.train_eval_split = train_eval_split
self.save_dir = save_dir
def generate_dataset(self):
for num_brick in tqdm(
range(self.min_num_brick, self.max_num_brick + 1),
"Number of Brick Progress",
):
# separate dataset for GN-based physics engine (GN Physics)
dataset_name = self.dataset_prefix + "_" + str(num_brick)
nodes_train, nodes_eval = [], []
edges_train, edges_eval = [], []
node_feats_train, node_feats_eval = [], []
edge_feats_train, edge_feats_eval = [], []
split = int(self.num_episode * self.train_eval_split)
for episode in tqdm(range(self.num_episode), "Episode Progress"):
if episode < split:
nodes_dummy = nodes_train
edges_dummy = edges_train
node_feats_dummy = node_feats_train
edge_feats_dummy = edge_feats_train
episode_dummy = copy.deepcopy(episode)
else:
nodes_dummy = nodes_eval
edges_dummy = edges_eval
node_feats_dummy = node_feats_eval
edge_feats_dummy = edge_feats_eval
episode_dummy = copy.deepcopy(episode) - split
# use test_simulator.py for debugging and visualization
p.connect(p.DIRECT)
p.setAdditionalSearchPath(pybullet_data.getDataPath())
p.setPhysicsEngineParameter(numSolverIterations=10)
p.setTimeStep(1.0 / 60.0)
# generate new design
self.designer.clear()
self.designer.generate_design(num_brick=num_brick)
for b in self.designer.built:
pos = b.anchor
rot = p.getQuaternionFromEuler(np.array(b.rotation) * np.pi / 2)
p.loadURDF("urdf/brick.urdf", pos, rot)
p.loadURDF("plane.urdf", useMaximalCoordinates=True)
# start simulation
p.setGravity(0, 0, -10)
for frame in range(self.num_frame):
p.stepSimulation()
time.sleep(1.0 / 240.0)
# remember the ground!
for objectID in range(num_brick + 1):
pos, ort = p.getBasePositionAndOrientation(objectID)
rot = p.getEulerFromQuaternion(ort)
vel_linear, vel_angular = p.getBaseVelocity(objectID)
if objectID < num_brick:
mass = 1.0
else:
mass = 9999999.0
nodes_dummy.append([episode_dummy, frame, objectID])
node_feats_dummy.append(
pos + rot + vel_linear + vel_angular + (mass,)
)
contact_list = p.getContactPoints()
for contact in contact_list:
edges_dummy.append(
[episode_dummy, frame, contact[1], contact[2]]
)
edge_feats_dummy.append(
contact[5] # positionA
+ contact[6] # positionB
+ (
contact[9], # normalForce
contact[10], # friction1
contact[12], # friction2
)
)
p.disconnect()
np.savez(
self.save_dir / str(dataset_name + "_TRAIN"),
nodes=np.array(nodes_train),
node_feats=np.array(node_feats_train),
edges=np.array(edges_train),
edge_feats=np.array(edge_feats_train),
)
np.savez(
self.save_dir / str(dataset_name + "_EVAL"),
nodes=np.array(nodes_eval),
node_feats=np.array(node_feats_eval),
edges=np.array(edges_eval),
edge_feats=np.array(edge_feats_eval),
)
if __name__ == "__main__":
parser = argparse.ArgumentParser(
"GN Based Physics Engine Dataset Creator For Lego Player"
)
parser.add_argument(
"--brick-type",
type=str,
default="plain",
help="Type of brick, either 'plain' or 'lego'",
)
parser.add_argument(
"--arena-length",
type=float,
default=1.0,
help="Arena length for designer to contain width of generated design",
)
parser.add_argument(
"--num-episode",
type=int,
default=200,
help="Number of episode per each separate dataset",
)
parser.add_argument(
"--num-frame",
type=int,
default=5,
help="Number of frame to run in simulator per each episode",
)
parser.add_argument(
"--min-num-brick",
type=int,
default=5,
help="Minimum number of bricks for separate dataset generation",
)
parser.add_argument(
"--max-num-brick",
type=int,
default=6,
help="Maximum number of bricks for separate dataset generation",
)
parser.add_argument(
"--brick-extents",
nargs="+",
type=float,
default=[0.4, 0.2, 0.1],
help="Brick extents for designer",
)
parser.add_argument(
"--safe-margin",
type=float,
default=0.1,
help="Safe margin for designer in deciding brick placement",
)
args = parser.parse_args()
if args.brick_type == "plain":
safe_margin = args.safe_margin
is_modular = False
mod_unit = None
rotate_prob = [0.1, 0.1, 0.5]
dataset_prefix = "BRICK_GNPHYSICS_PLAIN"
save_dir = pathlib.Path(__file__).parent.absolute() / "datasets" / "PLAIN"
elif args.brick_type == "lego":
is_modular = True
mod_unit = min(args.brick_extents)
rotate_prob = [0.0, 0.0, 0.5]
safe_margin = round(args.safe_margin / mod_unit) * mod_unit
dataset_prefix = "BRICK_GNPHYSICS_LEGO"
save_dir = pathlib.Path(__file__).parent.absolute() / "datasets" / "LEGO"
desigher = Designer(
arena_length=args.arena_length,
brick_extents=args.brick_extents,
safe_margin=safe_margin,
is_modular=is_modular,
mod_unit=mod_unit,
rotate_prob=rotate_prob,
)
dc = GNPhysicsDatasetCreator(
desigher,
dataset_prefix,
num_episode=args.num_episode,
num_frame=args.num_frame,
min_num_brick=args.min_num_brick,
max_num_brick=args.max_num_brick,
save_dir=save_dir,
)
dc.generate_dataset()