scope of research, background intro

thesis/Main.tex

@@ -194,11 +194,25 @@
     %\todo[inline, color=green!40]{contribution/idea of this thesis is to calculate a confidence score which describes how trustworthy input data is. algorithms further down the pipeline (slam, navigation, decision) can use this to make more informed decisions - examples: collect more data by reducing speed, find alternative routes, signal for help, do not attempt navigation, more heavily weight input from other sensors}
 
     \newsection{scope_research}{Scope of Research}
-    \todo[inline]{output is score, thresholding (yes/no), maybe confidence in sensor/data? NOT how this score is used in navigation/other decisions further down the line}
-    \todo[inline]{Sensor degradation due to dust/smoke not rain/fog/...}
-    \todo[inline, color=green!40]{we look at domain of rescue robots which save buried people after earthquakes, or in dangerous conditions (after fires, collapsed buildings) which means we are mostly working with indoors or subterranean environments which oftentimes are polluted by smoke and a lot of dust, ideally works for any kind of sensor data degradation but we only explore this domain}
-    \todo[inline, color=green!40]{mostly use lidar (state of the art) since they are very accurate in 3d mapping environments, so we focus on quantifying how trustworthy the lidar data is by itself. we do not look at other sensor data (tof, ultrasound, optical)}
-    \todo[inline, color=green!40]{intended output is confidence score which simply means higher score = worse data quality, lower score = trustworthy data. this score can be interpreted by algorithms in pipeline. we do not look at how this is implemented in the algorithms, no binary classifier but analog value, if this is wished followup algorithm has to decide (example by threshold or other methods)}
+    %\todo[inline]{output is score, thresholding (yes/no), maybe confidence in sensor/data? NOT how this score is used in navigation/other decisions further down the line}
+    %\todo[inline]{Sensor degradation due to dust/smoke not rain/fog/...}
+    %\todo[inline, color=green!40]{we look at domain of rescue robots which save buried people after earthquakes, or in dangerous conditions (after fires, collapsed buildings) which means we are mostly working with indoors or subterranean environments which oftentimes are polluted by smoke and a lot of dust, ideally works for any kind of sensor data degradation but we only explore this domain}
+
+    %In this thesis we limit the domain of our research to that of autonomous rescue robots and their unique challenges. The degradation of sensor data in this domain appears to mainly stem from airborne particles, which we evaluate. Other kinds of degradation like from adverse weather effects, specific material properties, irrationally moving structures like leaves from trees and others are out of scope of our research due to the low likelihood of them occuring in the rescue scenarios of autonomous robots. While our approach does not specifically exclude these types of degradation and actually a case can be built for the employed method allowing for quantifying any of these and more kinds of degradation, we do not explicitely look at them or evaluate any of them other than airborne particles.
+
+    In this thesis, we focus our research on the unique challenges faced by autonomous rescue robots, specifically the degradation of sensor data caused by airborne particles. While degradation in sensor data can also arise from adverse weather, material properties, or dynamic elements such as moving leaves, these factors are considered less relevant to the rescue scenarios targeted by our study and are therefore excluded. Although our method is versatile enough to quantify various types of degradation, our evaluation is limited to degradation from airborne particles, as this is the most prevalent issue in the operational environments of autonomous rescue robots.
+
+    %\todo[inline, color=green!40]{mostly use lidar (state of the art) since they are very accurate in 3d mapping environments, so we focus on quantifying how trustworthy the lidar data is by itself. we do not look at other sensor data (tof, ultrasound, optical)}
+
+    %While computer vision systems of robots oftentimes include a multitude of sensor types like time of flight cameras, IR cameras, ultrasound sensors and others, we found that for autonomous robots in rescue missions mapping and navigation challenges are so hard that they require and mostly rely on lidar sensor data which is very accurate, high resolution and allow for mapping the whole surroundings due to their high field of view which oftentimes contains the whole 360° horizontal fov and a quite large vertical fov as well. additionally the cost of these lidar sensors has plummeted over the last decades due to their use in autonomous driving, drones and robotics as well as advancements in manufacturing like the utilisation of microeletromechanical systems which makes their proliferation for these types of vision problems near universal. for these reasons we limit our research to data produced by lidar sensors, namely the pointclouds of measurements in a coordinate system they produce. oftentimes sensor data is fused to achieve better accuracy and higher confidence in data but firstly examining these sensor fusion data would most likely increase the initial complexity of the research too much and secondly it would limit our research to platforms which utilise all the sensors we included instead of being able to quantify the sensor degradation by the lidar data itself.
+
+    While robotic computer vision systems often incorporate a variety of sensors—such as time-of-flight cameras, infrared cameras, and ultrasound sensors—we found that autonomous rescue robots primarily depend on LiDAR data for mapping and navigation. LiDAR sensors offer high accuracy, high resolution, and an extensive field of view (often a full 360° horizontally and a substantial vertical coverage), which are essential for constructing comprehensive environmental maps in challenging scenarios. Furthermore, the cost of LiDAR sensors has decreased significantly in recent decades, driven by their widespread adoption in autonomous driving, drones, and robotics, as well as manufacturing advancements like microelectromechanical systems (MEMS). For these reasons, our research is focused exclusively on LiDAR sensor data—specifically, the point clouds generated within a defined coordinate system. Although sensor fusion techniques are commonly used to enhance data accuracy and confidence, incorporating fused data would not only add significant complexity to our study but also limit our analysis to platforms equipped with all the sensor types involved. Consequently, we concentrate on quantifying sensor degradation solely through LiDAR data.
+
+    %\todo[inline, color=green!40]{intended output is confidence score which simply means higher score = worse data quality, lower score = trustworthy data. this score can be interpreted by algorithms in pipeline. we do not look at how this is implemented in the algorithms, no binary classifier but analog value, if this is wished followup algorithm has to decide (example by threshold or other methods)}
+
+  %The output of the method utilized by us and in our experiments is an analog score which relates to the confidence in the data and inversely to the data degradation. We do not look at how such a score might be utilized but can see many applications like simple thresholding and depending on the outcome deciding not to proceed in a certain direction or a direct usage by for example tying the robots speed to its confidence in the data, if necessary slowing down and collecting more data before progressing. The output score is independent of the time dimension and just a snapshot of each lidar scans degradation. In reality many lidars produce multiple scans per second, which would allow for including the time series of data and the scores they produce into an analysis as well such as for example a running average of the score or more complex statistical analysis. We do not investigate the differences between lidar scans which were taken with a small time delta, only at single snapshots of data in time.
+
+  The method we employ produces an analog score that reflects the confidence in the sensor data, with lower confidence indicating higher degradation. Although we do not investigate the direct applications of this score, potential uses include simple thresholding to decide whether to proceed with a given action or dynamically adjusting the robot's speed based on data quality—slowing down to collect additional data when confidence is low. Importantly, this output score is a snapshot for each LiDAR scan and does not incorporate temporal information. While many LiDAR sensors capture multiple scans per second—enabling the possibility of time-series analyses such as running averages or more advanced statistical evaluations—we focus solely on individual scans without examining the differences between successive scans.
 
     \newsection{thesis_structure}{Structure of the Thesis}
     \todo[inline]{brief overview of thesis structure}
@@ -208,11 +222,20 @@
 
     %\todo[inline, color=green!40]{in this section we will discuss necessary background knowledge for our chosen method and the sensor data we work with. related work exists mostly from autonomous driving which does not include subter data and mostly looks at precipitation as source of degradation, we modeled after one such paper and try to adapt the same method for the domain of rescue robots, this method is a semi-supervised deep learning approach to anomaly detection which we describe in more detail in sections 2.1 and 2.2. in the last subsection 2.3 we discuss lidar sensors and the data they produce}
 
-  As the domain of robotics and embedded systems often does, this thesis constitutes quite  broad interdisciplinary challenge of various fields of study. As we will see in this chapter, anomaly detection-the methodology we posed our degradation quantification problem as-has roots in statistical analysis and finds utility in many domains. As is the case for many fields of study, there has been success in incorporating learning based techniques-especially deep learning-into it to better or more efficiently solve problems anchored in interpretation of large data amounts. The very nature of anomalies often times makes their form and structure unpredictable, which lends itself to unsupervised learning techniques-ones where the training data is not assigned labels beforehand, since you cannot label what you cannot expect. These unsupervised techniques can oftentimes be improved by utilizing a small but impactful number of labeled training data, which results in semi-supervised methods. 
+  %As the domain of robotics and embedded systems often does, this thesis constitutes quite  broad interdisciplinary challenge of various fields of study. As we will see in this chapter, anomaly detection-the methodology we posed our degradation quantification problem as-has roots in statistical analysis and finds utility in many domains. As is the case for many fields of study, there has been success in incorporating learning based techniques-especially deep learning-into it to better or more efficiently solve problems anchored in interpretation of large data amounts. The very nature of anomalies often times makes their form and structure unpredictable, which lends itself to unsupervised learning techniques-ones where the training data is not assigned labels beforehand, since you cannot label what you cannot expect. These unsupervised techniques can oftentimes be improved by utilizing a small but impactful number of labeled training data, which results in semi-supervised methods. The method we evaluate for our task-Deep SAD-is a not only a semi-supervised deep learning approach but also employs an autoencoder in its architecture, a type of neural network architecture which has found widespread use in many deep learning applications over the last few years due to its feature extraction capability which solely relies on unlabeleld data. Its approach typically lends itself especially well to complex data for which feature extraction of conventional manual methods is hard to achieve, like the lidar data we are working with in this thesis. Lidar sensors measure the range from the sensor to the next reflective object for many angles simultaneously by projecting a laser in a specified direction and measuring the time it takes a reflected ray to return to the sensor. From the output angles of the rays and the measured travel time the sensor can construct a point cloud which is oftentimes dense enough to map out the sensors surroundings. In this chapter we will discuss these necessary technologies, give an overview of their history, use-cases and describe how it will be utilized in this thesis.
+
+  This thesis tackles a broad, interdisciplinary challenge at the intersection of robotics, embedded systems, and data science. In this chapter, we introduce the background of anomaly detection—the framework we use to formulate our degradation quantification problem. Anomaly detection has its roots in statistical analysis and has been successfully applied in various domains. Recently, the incorporation of learning-based techniques, particularly deep learning, has enabled more efficient and effective analysis of large datasets.
+
+Because anomalies are, by nature, unpredictable in form and structure, unsupervised learning methods are often preferred since they do not require pre-assigned labels—a significant advantage when dealing with unforeseen data patterns. However, these methods can be further refined through the integration of a small amount of labeled data, giving rise to semi-supervised approaches. The method evaluated in this thesis, DeepSAD, is a semi-supervised deep learning approach that also leverages an autoencoder architecture. Autoencoders have gained widespread adoption in deep learning for their ability to extract features from unlabeled data, which is particularly useful for handling complex data types such as LiDAR scans.
+
+LiDAR sensors function by projecting lasers in multiple directions simultaneously, measuring the time it takes for each reflected ray to return. Using the angles and travel times, the sensor constructs a point cloud that is often dense enough to accurately map its surroundings. In the following sections, we will delve into these technologies, review their historical development and use cases, and describe how they are employed in this thesis.
   
 
     \newsection{anomaly_detection}{Anomaly Detection}
 
+    Anomaly detection refers to the process of detecting unexpected patterns of data, outliers which deviate significantly from the majority of data which is implicitly defined as normal by its prevalence. In classic statistical analysis these techniques have been studied as early as the 19th century~\cite{anomaly_detection_history}. Since then, a multitude of methods and use-cases for them have been proposed and studied.  
+
+
     \todo[inline, color=green!40]{cite exists since X and has been used to find anomalous data in many domains and works with all kinds of data types/structures (visual, audio, numbers). examples healthcare (computer vision diagnostics, early detection), financial anomalies (credit card fraud, maybe other example), security/safety video cameras (public, traffic, factories).}
     \todo[inline, color=green!40]{the goal of these algorithms is to differentiate between normal and anomalous data by finding statistically relevant information which separates the two, since these methods learn how normal data typically is distributed they do not have to have prior knowledge of the types of all anomalies, therefore can potentially detect unseen, unclassified anomalies as well. main challenges when implementing are that its difficult to cleanly separate normal from anormal data}
     \todo[inline, color=green!40]{typically no or very little labeled data is available and oftentimes the kinds of possible anomalies are unknown and therefore its not possible to label all of them. due to these circumstances anomaly detection methods oftentimes do not rely on labeled data but on the fact that normal circumstances make up the majority of training data (quasi per defintion)}

thesis/bib/bibliography.bib

@@ -140,7 +140,7 @@
   volume = {18},
   url = {https://api.semanticscholar.org/CorpusID:14006440},
 },
-@article{anomaly_detection_1800s,
+@article{anomaly_detection_history,
   title = {XLI. On discordant observations},
   author = {Francis Ysidro Edgeworth},
   journal = {Philosophical Magazine Series 1},