SCI每部分必备例句

1.abstract

Map generation by a robot in a cluttered and noisy environment
is an important problem in autonomous robot navigation.

Localization, mapping, remote operation, maintenance, and
health and safety are identified as the main beneficiaries
from rapidly developing technologies, such as 3-D visualization,
augmented reality, energy autonomous sensor nodes,
distributed sensing, smart network protocols, and big data analytics.

mobile robot automation relies heavily
on technologies such as sensing, localization, mapping.

摘要第一句和第二句：
1.背景
Robust localization and mapping by a robot in a cluttered and noisy indoor environment
is an important problem in autonomous robot navigation.

Simultaneous Localization and Mapping (SLAM)-based mobile mapping is an efficient method
to construct an indoor map.
2.目的第三句
1.The proposed solution exploits…
2.This paper proposes a … method based on…
3.In this paper we propose a method for …
4.This paper proposes an SLAM algorithm based on …
5.This paper presents an … algorithms applied to … for indoor environments
6.This paper therefore proposes …
7.This paper, however, proposes …
8.This paper further proposes a new method to
This paper proposes a SlAM algorithm based on Monte-Carlo Localization and Mapping
methods with discrete hough transform in order to build a robust robot navigation system.
3.方法第四句
1.This article is an attempt to…
2.
A grid-based occupancy likelihood map is chosen as the map representation method and is built from all previous scans.
The set of relative rotation angles and corresponding spatial displacements of the robot are estimated according to Hough spatial energy spectrum correlation functions construction of global and local 2D Occupancy Grid Maps,
Further more,to use the result serve as the guiding particle set of the Monte Carlo global location algorithm.
4.结果概述第五局
1.Several systematic experiments for evaluating the approach have been performed both
on a simulator and on soccer robots embedded in the RoboCup environment.
2.Experimental results are presented to illustrate the performance of the proposed algorithm.
3.We also present empirical results from experimental evaluations of our approach
in real world scenarios both from the perspective of the navigation capability and the usability of the interface.
5.Several systematic experiments for evaluating the proposed algorithm for mapping and localization
have been performed in both simulated and real world scenarios,
and found it efficiency and build maps with good self-localization.

Introduction
第一部分：研究背景及其重要性（有大到小，有古道今）
In the past 30 years, plenty of scan matching methods have been developed.

It is important for a mobile robot to be able to autonomously navigate and localize 

itself in a known or unknown environment. A precise position estimation always serves as
the heart in any navigation systems, such as localization,map building or path planning.

One primary issue is how to accurately match sensed data against information in a priori map or information that has

been continuously collected. There are two common matching techniques that havebeen used in mobile robotics: point-based
matching and feature-based matching.
Simultaneous localization and mapping (SLAM) in unknown GPS-denied environments is a major challenge
for researchers in the field of mobile robotics. Many solutions for single-robot SLAM exist;

Autonomous navigation by a mobile robot in an unknown

environment is an important problem in mobile robotics.
The robot mainly perceives the outer world by utilizing
measurement sensors attached to it. These sensors (typi‐
cally laser range finders, ultrasonic sensors, cameras and
3D sensors) give an estimate of the robot‘s position in the
environment and incrementally build the map of the
environment. This problem is often referred to as ’simulta‐
neous localization and mapping’ (SLAM), in which a robot
simultaneously localizes its position in the environment
and builds a map of it. Hence, it becomes important for the
robot to have an accurate map. While sensors such as wheel
encoders measure the relative robot displacement, external
sensor data can be used to correlate subsequent robot
positions to get the relative pose or estimate, improving
upon the odometry data. However, in reality, sensors are
prone to errors and generate noise in the data. This noise
then accumulates as error over time and gives the wrong
estimate or else the wrong map. The environments in which
the robot navigates are mostly dynamic, which adds to the
complexity of the mapping. The main task of the robot is to
incrementally build the map of the workspace as it discov‐
ers the environment, and to avoid obstacles, in order to
reach its final goal. Various approaches to solving this
problem have been discussed by researchers

Localization and navigation are essential problems in 

mobile robotics, which have been studied by many researchers
in this area [1-3]. Moreover, the accuracy of localization have
a direct influence on the accuracy of navigation, so a mobile
robot which can perform assigned task quickly and rightly is
based on high accuracy of localization.

Localization is a well studied problem in mobile robotics

since the information about the robot’s pose is essential for
many applications. Over the last 20 years, several probabilis-
tic localization techniques have been proposed and they have
demonstrated the capability to robustly estimate the pose of
robots in a large variety of application scenarios.

In recent research, ...  In our method, ...
The key contribution of this paper is a thorough


Establishing an accurate and clear indoor map is a basic requirement of Indoor Navigation and

Location-Based Services (INLBS). Mobile mapping is a highly efficient method for constructing an
indoor map. Simultaneous Localization and Mapping (SLAM) is a popular and applicable method for
mobile mapping in a GNSS-denied area, especially indoor environments [1,2].
SLAM estimates the pose and map in an unknown environment simultaneously [3]. According
to the type of sensors, the SLAM techniques can be classified into vision-based and range-based
approaches. Vision-based SLAM uses monocular, stereo or RGBD cameras to accomplish navigation
and mapping and can obtain rich texture information [4–8]. While it is the cheapest approach, there
are some disadvantages that limit its applications in indoor mapping, such as its sensitivity to lighting
conditions,computationalcostforprocessinglargeamountsofimagedataandthenecessityforaccurate
calibration [9]. Range-based SLAM using sonar or laser sensors are applied in most indoor mapping
techniques [9–12], because of its advantages of low complexity of data acquired and insensitivity to
lighting conditions [9]. In this paper, Light Detection and Ranging (LiDAR) is utilized, which is one
of the most popular ranging sensors, with a high ranging accuracy for SLAM [13,14], and which is
utilized in many SLAM-based mapping applications [15,16].
There are two different major map representation methods with different positioning approaches.
One of them is a feature-based map, in which the features of the indoor environment mainly
include points and line segments. Point features can be detected by vision sensors, but for LiDAR,
the line segment features are more detectable [1,9]. The positioning is completed using feature
matching [1,17–24]. Feature extraction and association are the important steps for feature matching,
which increases the complexity and may result in errors. Moreover, there may be no features in a
laser scan. Compared to feature matching, the scan matching utilizes two or more consecutive frames
of scan points directly for positioning. The classical scan matching algorithm is classical Iterative
Closed Point (ICP) [25]. There are many improved algorithms, based on ICP, such as Polar Scan
Matching (PSM) [26] and Iterative Closed Line (ICL) [27,28]. These methods calculate the relative
position of consecutive pairs of scans, but the errors will accumulate rapidly over time. To overcome
this problem, a grid-based occupancy likelihood map, which is an alternative method, is utilized to
store the historical data, and scan-to-map matching is proposed to find the optimal position [29–35].
The key to this approach is how it generates the gird-based occupancy likelihood map. Scan-to-map
matching is one of the most popular methods for indoor positioning and mapping, such as Hector
SLAM [32] and Google’s Cartographer [34]. One of the scan-to-map matching algorithms with an
occupancy likelihood map is Maximum Likelihood Estimation (MLE) [11,13,14,29,30]. In our work,
we utilize MLE for scan-to-map matching.
While scan-to-map matching can achieve positioning directly, without feature extraction and
association, some non-geometric or noise scan points produced by the objects, such as glass cabinets
and bookcases, will affect the process involved in finding an optimal rigid-body transformation and
decreasetheaccuracy. Moreover,thegeneralizedoccupancylikelihoodmapcontainsalltheinformation
collected by sensors, which includes useful structure information and noise information. It cannot
clearly represent the indoor environment. In artificial environments, line segments are the principal
elements and can provide considerable geometric information about the indoor environment. Using
line features for positioning, the accuracy of localization will be enhanced [19]. Kuo et al. proposed
a hybrid approach to enhance grid mapping by line matching and, for SLAM, utilized scan-to-map
matching and line feature matching together [19]. Feature matching and scan matching still work
independently. In fact, the line feature information could be utilized in generalizing the occupancy
likelihood map. Then, scan-to-mapping can be enhanced, without extra computing complexity.
In the approaches using line segments, the parameters and covariance matrix of the line segments
are calculated by the points that constitute a line segment [17–23]. This process compresses the data
information of the scan points into the line parameter, which increases the computing complexity and may
introduce errors, when the line segments are not very standard. It is better to use the points that constitute
a line segment directly for mapping. The lines in most indoor environments are considered parallel or
orthogonal to each other. This constraint is exploited to select line segments using various approaches in
order to improve the robustness of the orientation and filter out many dynamic objects [21–23].
Inertial Measurement Units (IMUs) have been applied in many SLAM systems, primarily for
attitude estimation [35–37]. The accurate attitude and orthogonal constraint can be combined to extract
the orthogonal line segments efficiently. In this paper, a 2-D LiDAR and a commercial-grade IMU are
utilized for 2D indoor mapping. An IMU-aided scan matching method is proposed for an orthogonal
weighted occupancy likelihood map (OWOLM).
Compared with existing LiDAR-based indoor mapping solutions, this paper offers two major
contributions. First, a new line segment extraction method is proposed, based on attitude-aided
blurred segments, and this method combines the attitude and orthogonal constraint to extract the orthogonal quickly,
without calculating the parameters of the lines. Second, the blurred segments are
given a higher weight in generalizing the occupancy likelihood map, which can effectively reduce
accumulated errors and construct a clearer indoor map. The remainder of this paper is organized as
follows: Section 2 describes the methods utilized in this research; Section 3 introduces the indoor field
tests and discusses the experimental results; and Section 4 draws conclusions.

The remainder of the paper is organized as follows. In 

Section II, a vision system based on QR codes is presented.
The localization method using QR codes are addressed in
details in Section III. Section IV discusses the experimental
results. Finally, Section V summarizes the paper and provides
future work.

Mobile robot localization is a well studied field in robotics

and several approaches to localization have been proposed
in the past [1], [2], [7], [10], [15], [19]. Probabilistic ap-
proaches have been successfully applied to localize robots
with respect to a given map in a robust manner, often relying
on techniques such as extended Kalman filters (EKF) [19],
histogram filters [13] or particle filters, often referred to
as Monte-Carlo localization (MCL) [8].

Localization systems are an essential enabling component

of mobile robotic systems. The importance of accurate posi-
tioning is even higher in modern logistics application scenar-
ios, which rely on precise and repeatable trajectory follow-
ing using Automatically Guided Vehicles (AGV). An AGV
typically has to achieve positioning accuracy of under 3cm,
especially during loading operations. Moreover, as AGVs are
typically non-holonomic platforms, sufficient accuracy has to
be maintained throughout the trajectory in order to guarantee
accuracy in orientation at loading points. In addition to being
accurate, the pose estimate must be smooth and updated in
real-time in order to avoid violent corrective control actions
which can be catastrophic for heavily loaded vehicles. As
a consequence of hard requirements, industry is still using
positioning methods that rely on fixed infrastructure in the
factories. Setting up a localization infrastructure is, however,
a high-cost investment and thus there is a strong demand for
more flexible solutions.
Monte Carlo Localization (MCL) [3] is among the most
popular localization approaches in the robotics research
community. MCL is a probabilistic map based localization
approach that has been shown to be robust in real-world
scenarios [3], [15], [9]. However, in prior contributions MCL
has not been reported to reach the precision requested by
industrial applications [3], [15], [9]. Röwekämper et al. in [9]
analyzed the accuracy of an MCL localization system with
KLD-sampling and reported mean MCL errors of approx.
6cm in a repeatability test. Their solution for improving the
accuracy was to perform scan matching alignment to reach
accurately the final pose. Unfortunately this approach only
works for holonomic vehicles, while the majority of AGVs
are non-holonomic and cannot reliably correct their position
and orientation over a short trajectory.
The insufficient accuracy of the standard MCL algorithm
is further confirmed by our experiments. Fig. 1a visualizes
the positioning error of a standard MCL approach. The figure
is composed of fifteen overlaid pictures recorded with a
stationary camera during a positioning accuracy test. In this
experiment the vehicle was instructed to repeatedly execute a
pre-determined trajectory and stop at the same location at the
end of each trial. The blurriness of Fig. 1a demonstrates that
even in a simple stationary environment the standard MCL
approach fails to meet the accuracy requirements needed in
typical industrial applications. We argue that the main reason
for this failure stems from the grid-based representation used
by MCL. Occupancy grids [8], used for map representation in
the standard MCL approach, are discretized by definition and
thus cannot represent the environment accurately, especially
in presence of sensor noise.
In this paper we propose to use Normal Distributions
Transform (NDT) representation for an MCL implementa-tion.
The NDT is a piecewise continuous representation,
which represents the space as a set of normal distributions
[1], [6]. It is a compact representation allowing to model
the environment with a significantly lower number of cells
than occupancy grid maps. Moreover, NDT is by nature
a likelihood model, which makes it an ideal candidate to
be utilized in MCL. The accuracy of the proposed Normal
Distributions Transform based Monte Carlo Localization
(NDT-MCL) is demonstrated in Fig. 1b, which is obtained
through the same procedure as for the grid-based MCL (Fig
1a).
The main contribution of this paper is to formulate NDT-
MCL and to evaluate it using both pre-recorded datasets and
in a closed-loop fashion on board of an AGV test platform.
The experiments performed confirm that NDT-MCL provides
accuracy and repeatability superior to those of the standard
grid-based MCL and comparable to the performance of a
commercial infrastructure based system.
The rest of this paper is organized as follows. Sec. 2
first introduces and formalizes the standard MCL approach,
before presenting the proposed NDT-MCL algorithm. Sec. 3
describes the test setup and analyzes the obtained results and
finally Sec. 4 concludes the paper with discussion.

2.Introduction

Intruction
第一部分：
研究背景和重要性
第一段：简述
It is important for a mobile robot to be able to autonomously navigate and localize
itself in a known or unknown environment. A precise position estimation always serves as
the heart in any navigation systems, such as localization,map building or path planning.
Simultaneous localization and mapping (SLAM) in unknown GPS-denied environments is a major challenge
for researchers in the field of mobile robotics.Establishing an accurate and clear indoor map is a basic
requirement of Indoor Navigation and Location-Based Services. Mobile mapping is a highly
efficient method for constructing an indoor map. Simultaneous Localization and Mapping (SLAM)
is a popular and applicable method for mobile mapping in a GNSS-denied area, especially indoor environments.
第二段：从古到今论述（slam,location,mapping）
Localization and mapping is a well known problem in mobile robotics
since the information about the robot’s pose is essential for
many applications. Over the last 30 years, plenty of SlAM algorithm have been developed.
several probabilistic localization techniques have been proposed and they have
demonstrated the capability to robustly estimate the pose of
robots in a large variety of application scenarios.According to the type of sensors,
the SLAM techniques can be classified into vision-based and range-based approaches.

%%%%「General introduction」:
For decades, one of the most popular ideas in … literature is the idea that …

Recent theoretical developments have revealed that …

A common strategy used to study … is …

These approaches have been influential in field … because of …

Most of the theories of … are however focused on explaining …

The field has gradually broadened as …

… has been widely adopted in the field of …

One of the major topics to be investigated in … is …

… is widely considered at the time of …

… phenomenon has been widely observed through…
%%%%「Problem definition」:
There is a further problem with …

One primary problem with … is that …

The foremost problems are the facts that …

A challenging problem which arises in this domain is …

A well-known problem with … is that it does not take into account the …

This appears as a more straightforward problem compared to the … turns out to be even more challenging because …

This poses some problems when carrying out the …

Unfortunately, this approach results in problems related to …

These constraints make the problem difficult to …

Most of the research in … field is aimed at solving …
%%%「Gaps in literature」:
There has been less previous evidence for …

Other studies have failed to …

To our knowledge, few studies have yielded …

Limited open literature to date has reported…

Only a few studies have shown that…

However, … has rarely been examined directly

Moreover, few studies have focused on …
%%「Problems solution」:
There are many alternative methods are available for solving…

A solution to this problem is proposed in …

One approach to solve this problem involves the use of … can be applied to overcome …

A number of works have shown that this problem can be solved by using …

A large number of alternative approaches have been developed to…

Potential techniques to solve contemporary problems in … are…

The standard solution to the problem is based on …

Broadly speaking, the problem can be addressed by …

Recently, a more general solution has been developed for…
%%「Study motivation」:
It is of interest to know whether … still …

It would be of special interest to…

We therefore analyzed … and investigated whether…

For this study, it was of importance to investigate …

We investigated whether … can be partly explained by …

To examine the impact of …, we …

One common approach to testify… is to…

A new approach is therefore needed for …
%%「Aims & objectives」:
The aim is to develop more sophisticated methods for …

For our first goal, we focus on two problems …

The aim here is to investigate …

This project aims to develop an overarching framework to…

The aim of the experiment is to compare …

The overall goal of this research was to pursue …

One of the major aims of this work was to create …

The main objective is to investigate methods for improving…

The objectives can be restated in the light of …

The objective is to demonstrate the feasibility of …
%%「Significance and advantages of your work」:
The aim is to develop more sophisticated methods for …

For our first goal, we focus on two problems …

The aim here is to investigate …

This project aims to develop an overarching framework to…

The aim of the experiment is to compare …

The overall goal of this research was to pursue …

One of the major aims of this work was to create …

The main objective is to investigate methods for improving…

The objectives can be restated in the light of …

The objective is to demonstrate the feasibility of …

discussion\conclusion and future work

reference \ref https://zhuanlan.zhihu.com/p/27538669
写完结果results之后，我们来归纳一下论文的discussion/conclusion。首先，看一下results 和conclusion的区别。

Result：侧重因某种原因所产生的最终结果。

Conclusion：强调仔细思考做出的推论或结论，或者某事的结局。

所以，result更侧重的是客观描述what，而conclusion就要包含作者的思考和探究，更侧重how和why。

因此，在这一部分主要是针对所研究的问题或者假设提出作者认为的答案，是对本篇论文阐述的内容以及所得出的发现做出高屋建瓴的解释, 以及对当前解决方法不足的讨论和对未来的展望。

按照一贯推崇的套路写法，这一部分也可以分为“四步走”。

第一步：在初始段落中，简要回顾一下目标或研究问题。然后，通过总结主要发现，介绍引言最后一段提出的问题的答案。具体讨论主要研究结果，因为它们涉及研究问题，而避免从结果部分重述数据的细节。这部分最多写六个句子：一句话重申研究的目的；一句话说明主要方法；总结主要发现的2-3句话；一句话说明主要结论。当然最好能用更为简练的语言的阐述，以三句为佳。

看这个例子：

We have shown that XXX methods can achieve an order-of-magnitude acceleration in hemodynamic response sampling by combining dense head coil array parallel data acquisition with distributed source modeling. (reiterating the purpose of the study重申研究的目的) Applying XXX to study the visual system using an event-related design, we found that the method is both sensitive and reliable, as demonstrated in the individual participant and group average results (stating the main approach + main findings 说明主要方法和发现). Localization accuracy was examined by comparing the XXX and conventional EPI task-related activity in primary visual cortex, with the comparison revealing a good spatial match (approximately 5 mm) between the two methods (main findings 主要发现). Our results in the visual cortex show a faster BOLD-contrast response compared to conventional EPI (stating the main conclusions 主要结论).

第二步：解释你的结果，为自己的答案提供支持和论据。讨论文献中根据已有信息得出的关键结果。引用相关作品，比较和解释你的发现。讨论任何意想不到的发现，并就该发现提出建议。陈述此项发现对于临床和/或科学含义。如果你需要总结出版文献的主要发现，请注意确保：这些文献质量很好（顶级期刊/英文刊物）；所提供的信息是具体的；从外部来源获得的所有信息都被适当和一致地引用；得出准确的结论。

看看例子：

Our results in the visual cortex show a faster BOLD-contrast response compared to conventional EPI. This is comparable to previous results (Janz et al., 1997; Pfeuffer et al., 2002). In comparison to previous studies, XXX methods exhibit two principal advantages……. (讨论文献中根据已有信息得出的关键结果。引用相关作品，比较和解释你的发现。)

We found that XXX acquisition and reconstruction had sufficient sensitivity to reliably detect task-related activity in primary visual areas with high temporal resolution; we also found in this study that we had reduced false positive activation compared to our first XXX attempts (Lin et al., 2006a). This is because the current implementation has better matching between the reference scan and the XXX scan: both scans used the same 3D RF excitation, frequency encoding, and phase encoding. The only difference is that the reference scan had partition encoding steps, but the XXX scans did not. The improved matching between the reference scans and the XXX acquisitions, as quantified… It helps to generate a better reconstruction with higher sensitivity and reduced false positive activation. (解释你的发现，如果有意想不到甚至是不尽如人意的结果， 都不要逃避。 讨论这些发现，并就该发现提出建议。陈述此项发现对于临床和/或科学含义)

这第二部分可以扩展，取决于你想讨论的内容。如果需要，也使用子标题。

第三步：在不影响本项研究有效性的前提下，对研究的局限性和弱点进行一段说明。讨论研究的局限性和未来可能的研究发现，其目的是解决目前研究存在的弱点。如果可以采用合适语言做出合理的实名，这样的讨论反而会增加文章可读性和文章的含金量。

However, D and D* were not equal to ADC b50–750 and ADC b0–50 in absolute value, respectively. This discrepancy may be attributed to the limitation imposed by the MR software used in this study in which the minimum b value was restricted to…… appropriate combination of b values for the separation of perfusion and diffusion must be investigated in detail. However, many clinical MR systems may limit the arbitrary setting used in the magnitude and/or the number of b values.

第四步：一个很简短的总结，将结论与本项研究的目标联系起来并进行强化。

结语部分，有些杂志可能需要单独的子标题（Conclusion）自成一节，写作时要注意具体要求。

这一部分主要是1-2句总结作者目前研究的主要发现：一句描述主要结论，一句描述潜在影响。特别注意的是，在没有可靠数据支持的情况下， 切不要夸大调查结果或给出不相关甚至是错误的结论。

We divided the normal ADC of human skeletal muscle into ADC b0–50 and ADCb50–750 based on the finding that there were two straight lines with different slopes in the SI vs. b-value logarithmic plots for skeletal muscle. Regardless of the muscle studied, the value of ADCb0–50, reflecting both diffusion and perfusion, was significantly higher than that of ADCb50–750, which approximates a true diffusion coefficient for skeletal muscle……Further research on the relationship of ADC with skeletal muscle architecture, fiber type, physiological function, metabolism, maturation level, aging and muscle injury is necessary for the application of DWI in clinical medicine and sports sciences.

以上的四步走就是写作Discussion/Conclusion部分的套路。

在结束之前还需要提醒大家一些常犯的错误。

讨论太长或太短，并且提出太多与当前研究无关的猜测；
太过广泛的文献综述；
讨论了没有在Results结果部分中提供的数据结果；
未讨论意外或差异性的发现，或夸大当前发现的重要性；