-
Notifications
You must be signed in to change notification settings - Fork 3
/
Copy pathdata.py
249 lines (208 loc) · 10.3 KB
/
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
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
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
#!/usr/bin/env python3
"""
Copyright (c) 2020, University of California, San Diego
All rights reserved.
Author: Jiangeng Dong <[email protected]>
Redistribution and use in source and binary forms, with or without
modification, are permitted provided that the following conditions are
met:
Redistributions of source code must retain the above copyright
notice, this list of conditions and the following disclaimer.
Redistributions in binary form must reproduce the above copyright
notice, this list of conditions and the following disclaimer in the
documentation and/or other materials provided with the distribution.
THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
"AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
(INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
"""
import yaml
import h5py
import torch
from typing import Any, Dict, Tuple
import numpy as np
from torch.utils.data import Dataset, DataLoader
from tqdm import tqdm
import sys
JOINT_RANGE = np.array([6.1083, 2.668, 3.4033, 3.194, 6.118, 3.6647, 6.118])
KITCHEN_TSR_RANGE = np.array([3.6086, 2.62952, 2.0635998, 6.284, 6.284, 6.284])
BARTENDER_TSR_RANGE = np.array([3.6086, 2.62952, 2.0635998, 6.284, 6.284, 6.284])
def load_train_data(env: str, use_text: bool, use_reach: bool, use_tsr_config: bool) -> Dict[str, np.ndarray]:
assert env in ["bartender", "kitchen"]
TSR_RANGE = KITCHEN_TSR_RANGE if env == "kitchen" else BARTENDER_TSR_RANGE
with open("data/dataset/description.yaml", "r") as f:
description = yaml.load(f, Loader=yaml.CLoader)
groups = description[env]["train"]
if env == "bartender":
objs = ["fuze_bottle", "juice", "coke_can", "plasticmug", "teakettle"]
else:
objs = ["fuze_bottle", "juice", "coke_can", "mugred", "mugblack", "pitcher", "door"]
f_voxel = h5py.File("data/dataset/{}_voxel.hdf5".format(env), "r")
f_path = h5py.File("data/dataset/{}_path.hdf5".format(env), "r")
if use_text:
f_task_embedding = h5py.File("data/dataset/{}_text_embedding.hdf5".format(env), "r")
else:
f_task_embedding = h5py.File("data/dataset/{}_ntp_embedding.hdf5".format(env), "r")
f_tsr = h5py.File("data/dataset/{}_tsr_path.hdf5".format(env), "r")
# calculate total size and preallocate space
inputs_length = 0
config_width = 7 if not use_tsr_config else 13
voxels_length = 0
voxel_shape = (32, 32, 32) if env == "kitchen" else (33, 33, 33)
task_embedding_width = 4096 if use_text else 270
for group in groups:
paths = f_path["pick_place"][group]
for obj in objs:
T = paths[obj].shape[0] - 1
inputs_length += T
voxels_length += 1
if use_reach:
paths = f_path["reach"][group]
for obj in objs:
T = paths[obj].shape[0] - 1
inputs_length += T
voxels_length += 1
result = {
"inputs": np.zeros((inputs_length, 2*config_width), dtype=np.float32),
"outputs": np.zeros((inputs_length, config_width), dtype=np.float32),
"distances": np.zeros((inputs_length,), dtype=np.float32),
"voxel_idxs": np.zeros((inputs_length,), dtype=np.int32),
"task_embedding_idxs": np.zeros((inputs_length,), dtype=np.int32),
"voxels": np.zeros((voxels_length, *voxel_shape), dtype=np.float32),
"task_embeddings": np.zeros((voxels_length, task_embedding_width), dtype=np.float32),
}
inputs_offset = 0
voxels_offset = 0
for group in tqdm(groups, desc="Load dataset from disk"):
voxels = f_voxel[group]
# process pick_place path
paths = f_path["pick_place"][group]
task_embeddings = f_task_embedding["pick_place"][group]
tsrs = f_tsr["pick_place"]["config"][group]
distances = f_tsr["pick_place"]["distance"][group]
for obj in objs:
voxel = voxels[obj]
task_embedding = task_embeddings[obj]
path = np.divide(paths[obj], JOINT_RANGE)
tsr = np.divide(tsrs[obj], TSR_RANGE)
distance = distances[obj]
if use_tsr_config:
path = np.concatenate((path, tsr), axis=1)
else:
path = np.array(path)
assert path.shape[0] == tsr.shape[0] and tsr.shape[0] == distance.shape[0]
T = path.shape[0] - 1
goal = path[-1]
goals = np.tile(goal, (T, 1))
starts = path[:-1]
nexts = path[1:]
result["inputs"][inputs_offset:inputs_offset+T] = np.concatenate((starts, goals), axis=1)
result["outputs"][inputs_offset:inputs_offset+T] = nexts
result["distances"][inputs_offset:inputs_offset+T] = distance[:-1]
result["voxel_idxs"][inputs_offset:inputs_offset+T] = voxels_offset
result["task_embedding_idxs"][inputs_offset:inputs_offset+T] = voxels_offset
result["voxels"][voxels_offset] = voxel
result["task_embeddings"][voxels_offset] = task_embedding
voxels_offset += 1
inputs_offset += T
if use_reach:
paths = f_path["reach"][group]
task_embeddings = f_task_embedding["reach"][group]
tsrs = f_tsr["reach"]["config"][group]
distances = f_tsr["reach"]["distance"][group]
for obj in objs:
voxel = voxels[obj]
task_embedding = task_embeddings[obj]
path = np.divide(paths[obj], JOINT_RANGE)
tsr = np.divide(tsrs[obj], TSR_RANGE)
distance = distances[obj]
if use_tsr_config:
path = np.concatenate((path, tsr), axis=1)
else:
path = np.array(path)
assert path.shape[0] == tsr.shape[0] and tsr.shape[0] == distance.shape[0]
T = path.shape[0] - 1
goal = path[-1]
goals = np.tile(goal, (T, 1))
starts = path[:-1]
nexts = path[1:]
result["inputs"][inputs_offset:inputs_offset+T] = np.concatenate((starts, goals), axis=1)
result["outputs"][inputs_offset:inputs_offset+T] = nexts
result["distances"][inputs_offset:inputs_offset+T] = distance[:-1]
result["voxel_idxs"][inputs_offset:inputs_offset+T] = voxels_offset
result["task_embedding_idxs"][inputs_offset:inputs_offset+T] = voxels_offset
result["voxels"][voxels_offset] = voxel
result["task_embeddings"][voxels_offset] = task_embedding
voxels_offset += 1
inputs_offset += T
assert voxels_offset == voxels_length
assert inputs_offset == inputs_length
f_voxel.close()
f_path.close()
f_task_embedding.close()
f_tsr.close()
return result
def load_test_data(env: str, use_text: bool) -> Dict[str, Dict[str, Dict[str, torch.Tensor]]]:
assert env in ["bartender", "kitchen"]
with open("data/dataset/description.yaml", "r") as f:
description = yaml.load(f, Loader=yaml.CLoader)
groups = description[env]["test"]
if env == "bartender":
objs = ["fuze_bottle", "juice", "coke_can", "plasticmug", "teakettle"]
else:
objs = ["fuze_bottle", "juice", "coke_can", "mugred", "mugblack", "pitcher", "door"]
f_voxel = h5py.File("data/dataset/{}_voxel.hdf5".format(env), "r")
if use_text:
f_task_embedding = h5py.File("data/dataset/{}_text_embedding.hdf5".format(env), "r")
else:
f_task_embedding = h5py.File("data/dataset/{}_ntp_embedding.hdf5".format(env), "r")
result = {
"voxel": {},
"task_embedding": {}
}
for group in tqdm(groups, desc="Load test dataset from disk"):
voxels = f_voxel[group]
task_embeddings = f_task_embedding["pick_place"][group]
result["voxel"][group] = {}
result["task_embedding"][group] = {}
for obj in objs:
voxel = voxels[obj]
task_embedding = task_embeddings[obj]
result["voxel"][group][obj] = torch.from_numpy(np.array(voxel).astype(np.float32))
result["task_embedding"][group][obj] = torch.from_numpy(np.array(task_embedding).astype(np.float32))
f_voxel.close()
f_task_embedding.close()
return result
class CoMPNetXDataset(Dataset):
def __init__(self, env: str, use_text: bool, use_reach: bool, use_tsr_config: bool, use_manifold_distance: bool) -> None:
super().__init__()
data = load_train_data(env, use_text, use_reach, use_tsr_config)
self.inputs = data["inputs"]
self.outputs = data["outputs"]
self.distances = data["distances"] if use_manifold_distance else None
self.voxel_idxs = data["voxel_idxs"]
self.task_embedding_idxs = data["task_embedding_idxs"]
self.voxels = data["voxels"]
self.task_embeddings = data["task_embeddings"]
self.size = self.inputs.shape[0]
self.use_manifold_distance = use_manifold_distance
def __getitem__(self, index: int) -> Tuple[torch.Tensor, torch.Tensor, torch.Tensor, torch.Tensor, torch.Tensor]:
result = [self.inputs[index],
self.outputs[index],
self.voxels[self.voxel_idxs[index]],
self.task_embeddings[self.task_embedding_idxs[index]]] + ([self.distances[index]] if self.use_manifold_distance else [])
return tuple(torch.as_tensor(value) for value in result)
def __len__(self) -> int:
return self.size
if __name__ == "__main__":
np.set_printoptions(threshold=sys.maxsize)
result = load_train_data("bartender", use_text=False, use_reach=False, use_tsr_config=True)
print(np.max(result["outputs"] * np.concatenate((JOINT_RANGE, BARTENDER_TSR_RANGE)), axis=0))
print(np.min(result["outputs"] * np.concatenate((JOINT_RANGE, BARTENDER_TSR_RANGE)), axis=0))