-
Notifications
You must be signed in to change notification settings - Fork 0
/
scene3.html
239 lines (173 loc) · 8.05 KB
/
scene3.html
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
<!DOCTYPE html>
<html>
<head lang="en">
<meta charset="UTF-8">
<meta http-equiv="x-ua-compatible" content="ie=edge">
<title>Physically Grounded VLMs</title>
<meta name="description" content="Physically Grounded VLMs">
<meta name="viewport" content="width=device-width, initial-scale=1">
<!-- <base href="/"> -->
<!-- <link rel="apple-touch-icon" href="apple-touch-icon.png"> -->
<!-- <link rel="icon" type="image/png" href="img/seal_icon.png"> -->
<!-- Place favicon.ico in the root directory -->
<link rel="stylesheet" href="https://maxcdn.bootstrapcdn.com/bootstrap/3.3.5/css/bootstrap.min.css">
<link rel="stylesheet" href="https://maxcdn.bootstrapcdn.com/font-awesome/4.4.0/css/font-awesome.min.css">
<link rel="stylesheet" href="https://cdnjs.cloudflare.com/ajax/libs/codemirror/5.8.0/codemirror.min.css">
<link rel="stylesheet" href="css/app.css">
<link rel="stylesheet" href="css/bootstrap.min.css">
<script src="https://ajax.googleapis.com/ajax/libs/jquery/1.11.3/jquery.min.js"></script>
<script src="https://maxcdn.bootstrapcdn.com/bootstrap/3.3.5/js/bootstrap.min.js"></script>
<script src="https://cdnjs.cloudflare.com/ajax/libs/codemirror/5.8.0/codemirror.min.js"></script>
<script src="https://cdnjs.cloudflare.com/ajax/libs/clipboard.js/1.5.3/clipboard.min.js"></script>
<script src="js/app.js"></script>
<!-- Google tag (gtag.js) -->
<script async src="https://www.googletagmanager.com/gtag/js?id=G-52J0PM8XKV"></script>
<script>
window.dataLayer = window.dataLayer || [];
function gtag(){dataLayer.push(arguments);}
gtag('js', new Date());
gtag('config', 'G-52J0PM8XKV');
</script>
<style>
.nav-pills {
position: relative;
display: inline;
}
.imtip {
position: absolute;
top: 0;
left: 0;
}
</style>
</head>
<body>
<div class="topnav">
<a href="index.html">Home</a>
<div class="rightnav">
<a href="scene1.html">Scene 1: Countertop</a>
<a href="scene2.html">Scene 2: Art Table</a>
<a href="scene3.html" class="active">Scene 3: Floor</a>
<a href="scene4.html">Scene 4: Kitchen A</a>
<a href="scene5.html">Scene 5: Kitchen B</a>
<a href="scene6.html">Scene 6: Salad Bar</a>
<a href="scene7.html">Scene 7: Living Room</a>
<a href="scene8.html">Scene 8: Shelf</a>
<a href="RS1/index.html">Robot Scene 1</a>
<a href="RS2/index.html">Robot Scene 2</a>
</div>
</div>
<div class="container" id="main">
<div class="row">
<div class="col-md-8 col-md-offset-2">
<br>
<h1>
Scene 3: Floor
</h1>
<div class="irow">
<div class="strict_column">
<img src="img/eval2_scene3.png">
</div>
<div class="strict_column">
<p style="color: #ff00ff">A: clothing [green hoodie]</p>
<p>B: towel</p>
<p>C: clothing [striped shirt]</p>
<p>D: bottle [sunscreen bottle]</p>
<p style="color: #9900ff">E: towel [socks]</p>
<p>F: mouse [ear thermometer]</p>
<p>G: suitcase</p>
<p>H: bottle [hand sanitizer]</p>
<p>I: hair dryer [dumbbell]</p>
<p style="color: #4a86e8">J: clothing [blue shirt]</p>
</div>
</div>
<br>
<p class="text-justify">
Here we provide a scene from our real scene planning evaluation, and one example task for it. We provide the object detections and labels from OWL-ViT, which is what our planner has access to, but we provide more precise labels in square brackets. We provide color-coded bounding boxes for the object detections that are relevant for the example task. For each version of the LLM planner, we provide its output, including possible interaction with a VLM, and the generated plan. Outputs generated by the VLMs are in <span style="color: #ff9900">orange</span>, and we mark correct and incorrect plans with <span style="color: #38761d">green</span> and <span style="color: #ff0000">red</span> respectively.
</p>
<h2><b>Task 17: Bring me the lightest piece of clothing.</b></h2>
<p class="text-justify">
In this task, we want to determine if the VLM can reason about mass and maintain an understanding of what "clothes" are. Only <span style="color: #9900ff"> E </span> should be brought, since socks are lighter than any other piece of clothing in this picture.
</p>
<h3>Generated Plans</h3>
<div class="irow">
<div class="tcolumn">
<h3> No VLM </h3>
<p> Thought: I cannot determine the weight of the objects, but I will bring one piece of clothing to the human. </p>
<p> Plan: </p>
<p style="color: #ff0000">
1. Go to object A <br>
2. Pick up object A<br>
3. Bring to human object A <br>
4. Done </p>
</div>
<div class="tcolumn">
<h3> InstructBLIP </h3>
<p> Thought: I can ask which objects are clothing out of all the objects I can carry. Of these clothing items, I can ask which is the lightest. </p>
<p> Question about [A, B, C, D, E, F, G, H, I, J]: Is this object clothing? </p>
<p style="color: #ff9900"> Answer <br>
A: Yes (0.78), No (0.19), Unknown (0.01)<br>
B: Yes (0.85), No (0.13), Unknown (0.0)<br>
C: Yes (0.95), No (0.04), Clothing (0.0)<br>
D: No (0.9), Yes (0.09), Unknown (0.0)<br>
E: Yes (0.96), No (0.03), Clothing (0.0)<br>
F: No (0.8), Yes (0.16), Unknown (0.01)<br>
G: No (0.63), Yes (0.33), Luggage (0.01)<br>
H: No (0.72), Yes (0.23), Unknown (0.01)<br>
I: No (0.48), Yes (0.47), Unknown (0.02)<br>
J: Yes (0.84), No (0.14), Unknown (0.0)<br>
</p>
<p> Thought: The objects that are confidently clothing are A, B, C, E, and J. I should ask which is the lightest next. </p>
<p> Question about [A, B, C, E, J]: Is this object light? </p>
<p style="color: #ff9900"> Answer <br>
A: Yes (0.65), No (0.23), Unknown (0.09)<br>
B: Yes (0.69), No (0.2), Unknown (0.08)<br>
C: Yes (0.63), No (0.29), Unknown (0.05)<br>
E: Yes (0.76), No (0.18), Unknown (0.04)<br>
J: Yes (0.76), No (0.15), Unknown (0.06)<br>
</p>
<p> Thought: The lightest piece of clothing is the one with the highest Yes value or the lowest No value (if none are Yes). Since some of the answers are Yes, the answer is E (the highest yes value). </p>
<p> Plan: </p>
<p style="color: #38761d">
1. Go to object E <br>
2. Pick up object E <br>
3. Bring to human object E <br>
4. Done
</p>
</div>
<div class="tcolumn">
<h3> PG-InstructBLIP (ours) </h3>
<p> Thought: I can first ask which objects are clothing out of all the objects I can carry. Of these clothing items, I can ask which is the lightest. </p>
<p> Question about [A, B, C, D, E, F, G, H, I, J]: Is this object clothing? </p>
<p style="color: #ff9900"> Answer <br>
A: Yes (0.9), No (0.09), Fabric (0.0)<br>
B: Yes (0.84), No (0.14), Fabric (0.01)<br>
C: Yes (0.9), No (0.08), Clothing (0.01)<br>
D: No (0.93), Yes (0.06), Unknown (0.0)<br>
E: Yes (0.95), No (0.04), Clothing (0.0)<br>
F: No (0.95), Yes (0.04), Nothing (0.0)<br>
G: Yes (0.58), No (0.38), Bag (0.0)<br>
H: No (0.94), Yes (0.05), Unknown (0.0)<br>
I: No (0.75), Yes (0.24), Unknown (0.0)<br>
J: Yes (0.94), No (0.05), Fabric (0.0)<br>
</p>
<p> Thought: The objects that are confidently clothing are A, C, and J. I should ask which is the lightest next. </p>
<p> Question about [A, C, J]: Is this object light? </p>
<p style="color: #ff9900"> Answer <br>
A: No (0.6), Yes (0.35), Unknown (0.03)<br>
C: No (0.55), Yes (0.38), Unknown (0.05)<br>
J: Yes (0.51), No (0.45), Unknown (0.03)<br>
</p>
<p> Thought: The lightest piece of clothing is the one with the highest Yes value or the lowest No value (if none are Yes). Since some of the answers are Yes, the answer is J (the highest yes value). </p>
<p> Plan: </p>
<p style="color: #ff0000">
1. Go to object J <br>
2. Pick up object J <br>
3. Bring to human object J <br>
4. Done
</p>
</div>
</div>
</div>
</div>
</body>
</html>