-
Notifications
You must be signed in to change notification settings - Fork 0
/
scene1.html
206 lines (155 loc) · 6.99 KB
/
scene1.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
<!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" class="active">Scene 1: Countertop</a>
<a href="scene2.html">Scene 2: Art Table</a>
<a href="scene3.html">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 1: Countertop
</h1>
<div class="irow">
<div class="strict_column">
<img src="img/eval2_scene1.png">
</div>
<div class="strict_column">
<p style="color: #bf9000"> A: bottle </p>
<p style="color: #a64d79"> B: pitcher (container) </p>
<p style="color: #45818e"> C: bowl [flatter bowl] </p>
<p> D: towel [shirt] </p>
<p> E: countertop </p>
<p style="color: #9900ff"> F: bowl [taller ceramic bowl] </p>
<p> G: measuring cup [lock] </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 3: Bring me all containers that you can confidently determine have water.</b></h2>
<p class="text-justify">
In this task, we want to determine if the VLM makes assumptions regarding the contents of containers. For instance, <span style="color: #bf9000"> A </span> is a dark container, so we cannot tell if the liquid inside is water, and containers <span style="color: #45818e"> C </span> and <span style="color: #9900ff"> F </span> are empty. The only container that should be brought is <span style="color: #a64d79"> B </span>, which visibly has water inside.
</p>
<h3>Generated Plans</h3>
<div class="irow">
<div class="tcolumn">
<h3> No VLM </h3>
<p> Thought: I cannot confidently determine which containers have water without additional information. </p>
<p> Plan: </p>
<p style="color: #ff0000"> 1. Done </p>
</div>
<div class="tcolumn">
<h3> InstructBLIP </h3>
<p> Thought: I can ask which objects have water out of all the objects I can carry. I will only bring the ones with a confidence score over 0.8. </p>
<p> Question about [A, B, C, D, F, G]: Does this object have water? </p>
<p style="color: #ff9900"> Answer <br>
A: Yes (0.9), No (0.09), Unknown (0.0) <br>
B: Yes (0.99), No (0.01), Water (0.0) <br>
C: Yes (0.98), No (0.02), Water (0.0) <br>
D: No (0.67), Yes (0.25), Unknown (0.04) <br>
F: Yes (0.95), No (0.04), Water (0.0) <br>
G: No (0.8), Yes (0.15), Unknown (0.02)
</p>
<p> Thought: The objects with water and a confidence score over 0.8 are A, B, C, and F. </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. Put down object A<br>
5. Go to object B<br>
6. Pick up object B<br>
7. Bring to human object B<br>
8. Put down object B<br>
9. Go to object C<br>
10. Pick up object C<br>
11. Bring to human object C<br>
12. Put down object C<br>
13. Go to object F<br>
14. Pick up object F<br>
15. Bring to human object F<br>
16. Put down object F<br>
17. Done
</p>
</div>
<div class="tcolumn">
<h3> PG-InstructBLIP (ours) </h3>
<p> Thought: I can ask which objects have water out of all the objects I can carry. I will only bring the ones with a confidence score over 0.8. </p>
<p> Question about [A, B, C, D, F, G]: Does this object have water? </p>
<p style="color: #ff9900"> Answer <br>
A: Yes (0.58), No (0.39), Unknown (0.02) <br>
B: Yes (0.98), No (0.01), Water (0.0)<br>
C: No (0.96), Yes (0.03), Nothing (0.01)<br>
D: No (0.87), Yes (0.1), Unknown (0.01)<br>
F: No (0.93), Yes (0.06), Nothing (0.01)<br>
G: No (0.93), Yes (0.05), Unknown (0.01)
</p>
<p> Thought: The only object that has a confidence score over 0.8 for having water is B. </p>
<p> Plan: </p>
<p style="color: #38761d">
1. Go to object B <br>
2. Pick up object B <br>
3. Bring to human object B <br>
4. Done
</p>
</div>
</div>
</div>
</div>
</body>
</html>