-
Notifications
You must be signed in to change notification settings - Fork 0
/
lampilot.html
135 lines (117 loc) · 4.91 KB
/
lampilot.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
<!DOCTYPE HTML>
<!--
Massively by HTML5 UP
html5up.net | @ajlkn
Free for personal and commercial use under the CCA 3.0 license (html5up.net/license)
-->
<html>
<head>
<title>LaMPilot|Purdue Digital Twin Lab</title>
<meta charset="utf-8" />
<meta name="viewport" content="width=device-width, initial-scale=1, user-scalable=no" />
<link rel="stylesheet" href="assets/css/main.css" />
<link rel="icon" href="images/icon_36_36.png"/>
<noscript>
<link rel="stylesheet" href="assets/css/noscript.css" />
</noscript>
</head>
<body class="is-preload">
<!-- Wrapper -->
<div id="wrapper">
<!-- Header -->
<header id="header">
<a href="index.html" class="logo">LLM4AD</a>
</header>
<!-- Nav -->
<nav id="nav">
<ul class="links">
<li><a href="index.html">Overview</a></li>
<li><a href="talk2drive.html">Talk2Drive</a></li>
<!-- <li><a href="generic.html">Demos</a></li> -->
<li><a href="3R.html">3R</a></li>
<li><a href="survey.html">Survey</a></li>
<li class="active"><a href="lampilot.html">Lampilot</a></li>
</ul>
</nav>
<!-- Main -->
<div id="main">
<!-- Post -->
<section class="post">
<center>
<h2>LaMPilot: An Open Benchmark Dataset for Autonomous Driving with Language Model Programs<br />
</h2>
</center>
<center>
<h4>Yunsheng Ma*, Can Cui*, Xu Cao*, Wenqian Ye, Peiran Liu, Juanwu Lu, Amr Abdelraouf, Rohit Gupta,
Kyungtae Han, Aniket Bera, James M. Rehg and Ziran Wang<br /></h4>
</center>
<ul class="actions special">
<li><a href="https://arxiv.org/abs/2312.04372" class="button large">Paper</a></li>
</ul>
<p>A novel framework for planning in the field of autonomous driving, rethinking the task as a
code-generation process that leverages established behavioral primitives.<br /></p>
</header>
<header>
<p>Main Contributions:<br />
1. We introduce the first benchmark for evaluating LLM-based agents in autonomous driving
contexts. <br />
2. We propose a framework including an open simulator for executing policies, equipped with
scoring mechanisms.<br />
3. Our LaMPilot provides LLM agents with a set of APIs that cover essential driving functions,
where safety criteria are
loaded into the APIs. <br />
4. Our LaMPilot enables the translation of natural language instructions into actionable
strategies through code generation.<br />
<!-- <a href="#" class="image main"><img src="images/t2d.png" alt="" /></a> -->
<hr />
<h2>LaMPilot Framework Architecture</h2>
LaMPilot framework architecture. This diagram shows the workflows of the LaMPilot framework,
illustrating the process
from initialization to evaluation. Once initialized, a structured language generator interprets the
simulator
environment to establish the language context. The central LLM handles input prompts, utilizing
predefined APIs and
generating LMP. The LMP is then executed in the simulator. The executed policy is assessed by the
evaluator, which
provides a final score for the policy.
<div class="image main"><img src="images/lampilot.png" alt="" /></div>
<hr />
<header>
<h2>Autonomous driving function modules and message flow</h2>
</header>
<p> <span class="image left"><img src="images/freq.png" alt="" /></span>The dataset is divided into
three sets: training
(3,900 samples), validation (500 samples), and testing (500 samples). The dataset contains a
variety of instructions that reflect
the diversity and unpredictability of real-world requirements. These instructions are
categorized by maneuver types, such as turning,
lane changing, and overtaking, as well as scenario types, including both highway and
intersection settings. The right figure shows a
distribution of the first four words in the instructions, highlighting the diversity of
instructions in our dataset.
</p>
<!-- <div class="image main"><img src="images/pic01.jpg" alt="" /></div> -->
<hr />
</header>
<h2>Evaluation of Baselines on LaMPilot</h2>
<div class="image main"><img src="images/table_lampbase.png" alt="" /></div>
</div>
</div>
<!-- Copyright -->
<div id="copyright">
<ul>
<li>© Purdue DIgital Twin Lab</li>
<li>Lab Website: <a href="https://purduedigitaltwin.github.io">purduedigitaltwin.github.io</a></li>
</ul>
</div>
</div>
<!-- Scripts -->
<script src="assets/js/jquery.min.js"></script>
<script src="assets/js/jquery.scrollex.min.js"></script>
<script src="assets/js/jquery.scrolly.min.js"></script>
<script src="assets/js/browser.min.js"></script>
<script src="assets/js/breakpoints.min.js"></script>
<script src="assets/js/util.js"></script>
<script src="assets/js/main.js"></script>
</body>
</html>