ArticlePDF Available

Self-Tuning Method for Increased Obstacle Detection Reliability Based on Internet of Things LiDAR Sensor Models


Abstract and Figures

On-chip LiDAR sensors for vehicle collision avoidance are a rapidly expanding area of research and development. The assessment of reliable obstacle detection using data collected by LiDAR sensors has become a key issue that the scientific community is actively exploring. The design of a self-tuning methodology and its implementation are presented in this paper, to maximize the reliability of LiDAR sensors network for obstacle detection in the 'Internet of Things' (IoT) mobility scenarios. The Webots Automobile 3D simulation tool for emulating sensor interaction in complex driving environments is selected in order to achieve that objective. Furthermore, a model-based framework is defined that employs a point-cloud clustering technique, and an error-based prediction model library that is composed of a multilayer perceptron neural network, and k-nearest neighbors and linear regression models. Finally, a reinforcement learning technique, specifically a Q-learning method, is implemented to determine the number of LiDAR sensors that are required to increase sensor reliability for obstacle localization tasks. In addition, a IoT driving assistance user scenario, connecting a five LiDAR sensor network is designed and implemented to validate the accuracy of the computational intelligence-based framework. The results demonstrated that the self-tuning method is an appropriate strategy to increase the reliability of the sensor network while minimizing detection thresholds.
Content may be subject to copyright.
Sensors 2018, 18, 1508; doi:10.3390/s18051508
Self-Tuning Method for Increased Obstacle
Detection Reliability Based on Internet of Things
LiDAR Sensor Models
Fernando Castaño 1,*, Gerardo Beruvides 1, Alberto Villalonga 1,2, and Rodolfo E. Haber 1
1 Centre for Automation and Robotics, UPMCSIC, 28500 Arganda del Rey, Spain; (G.B.); (A.V.); (R.E.H.)
2 Research Centre of Advanced and Sustainable Manufacturing, UM, Matanzas 44100, Cuba
* Correspondence:; Tel.: +34-918-711-900
Received: 28 February 2018; Accepted: 8 May 2018; Published: 10 May 2018
Abstract: On-chip LiDAR sensors for vehicle collision avoidance are a rapidly expanding area of
research and development. The assessment of reliable obstacle detection using data collected by
LiDAR sensors has become a key issue that the scientific community is actively exploring. The
design of a self-tuning methodology and its implementation are presented in this paper, to
maximize the reliability of LiDAR sensors network for obstacle detection in the Internet of Things
(IoT) mobility scenarios. The Webots Automobile 3D simulation tool for emulating sensor
interaction in complex driving environments is selected in order to achieve that objective.
Furthermore, a model-based framework is defined that employs a point-cloud clustering technique,
and an error-based prediction model library that is composed of a multilayer perceptron neural
network, and k-nearest neighbors and linear regression models. Finally, a reinforcement learning
technique, specifically a Q-learning method, is implemented to determine the number of LiDAR
sensors that are required to increase sensor reliability for obstacle localization tasks. In addition, a
IoT driving assistance user scenario, connecting a five LiDAR sensor network is designed and
implemented to validate the accuracy of the computational intelligence-based framework. The
results demonstrated that the self-tuning method is an appropriate strategy to increase the reliability
of the sensor network while minimizing detection thresholds.
Keywords: LiDAR sensors reliability; Internet of Things; self-turning parameterization; k-nearest
neighbors; driven-assistance simulator
1. Introduction
Nowadays, Internet of Things (IoT) applications are present in many sectors that range from
industrial environments (e.g., manufacturing, energy, etc.) to our personal lives (e.g., health, society,
mobility, etc.). The IoT is a strategic innovation in automotive applications that has received support
and investment in recent years, in order to develop smart mobility ecosystems for the market with
an autonomous level of interaction between vehicles and infrastructures. Nevertheless, mainstream
car manufacturers, OEMs for the automotive sector, researchers, and engineers are introducing new
technological contributions as new short-term challenges have to be addressed [1,2].
One particular challenge of autonomous driving is the accuracy and reliability estimation in
vision devices such as light detection and ranging (LiDAR) sensors and stereo cameras integrated in
automotive driving assistance systems for pattern recognition and obstacle detection tasks [3]. In
many scenarios, it is very difficult to certify the real topology and distance of objects at a lower level
of uncertainty, in most cases due to dead zones, object transparency, light reflection, weather
Sensors 2018, 18, 1508 2 of 16
conditions, and sensor failures, among others [4]. Furthermore, traditional networking devices are
not designed for use in unpredictable, variable, and dynamic environments that characterize an IoT
transportation ecosystem, making it necessary to develop new methodologies that characterize and
estimate sensor reliability [5]. Sensor fusion is commonly applied to combine different sensors for
road detection, mainly cameras and LiDARs. Current sensor fusion methods are even taking
advantage of both types of sensors (cameras or LiDARs), rather than exploiting the advantages of
each individual type of sensor [6]. Furthermore, the parallel processing of frames (from a camera)
and (LiDAR) scans implies a high computational cost and is unnecessary in many scenarios where a
sensor-based error-prediction model for assessing runtime reliability is operative [7].
Another important issue is the increase of computing power and wireless communication
capabilities to expand the role of sensors from mere data collection to more demanding tasks that
include sensor fusion, classification, and collaborative target tracking. Fault tolerance and reliability
perform a key role for embedded systems, such as obscured wireless sensors that are deployed in
applications where physical access is difficult [8]. Reliable monitoring of a phenomenon (or event
detection) depends on the set of data provided by the cluster of sensors rather than any individual
node. The failure of one or more nodes may not cause the disconnection of operational data sources
from the data sinks (command nodes or end user stations). However, node failures may increase the
number of hops a data message has to go through before reaching its destination (thereby increasing
the message delay), providing an estimation of the failure probabilities of both the sensors and the
intermediate nodes (nodes used to relay messages between data sources, and data sinks) [9].
Several reconstruction methods are reported in the literature to create specific geometric models
of existing objects from scanned point clouds based on information obtained from LiDARs [10].
Progress with modelling techniques that simulate complex driving environments has led to realistic
representations of multiple input/output variables, which have been used to determine the most
influential factors in degraded reliability and to detect pedestrians, obstacles, and vehicles in real-
time driving scenarios [11]. Clustering techniques are intensively used in exploratory data mining,
statistical analysis, pattern recognition, image analysis, information retrieval, bioinformatics, data
compression, and computer graphics [12]. One widely-used clustering algorithm is the k-nearest
neighbor (kNN) algorithm, due to its effectiveness at isolating the nearest neighbors of a query in a
training dataset and then predicting the query with the major class of nearest neighbors [13]. Another
widely applied technique in industrial applications is reinforcement learning [14]. Exemplary results
have been achieved with the Q-learning algorithm [15], which generates artificial intelligence and
self-learning strategies in complex processes and that has a self-tuning capability to obtain the
optimal configuration, based on rewards or penalties learned in previous states (iterative knowledge
generation) [16].
The main contributions of this work are the design a four-step method and its implementation
to maximize the reliability of an IoT LiDAR sensor network and to minimize the detection threshold
(γ, the number of LiDAR sensors required to detect one obstacle). The method includes point cloud
grouping, modeling, learning, and self-tuning (knowledge-based learning algorithm) tasks,
combining supervised and reinforcement machine learning techniques and clustering. Furthermore,
an IoT driving assistance scenario with a sensor network is created using the Webots simulation tool
to generate a LiDAR scan benchmark. Finally, the method is validated in a dynamic obstacle
detention scenario, in order to obtain the best prediction model and the optimal number of LiDAR
sensors needed to guarantee obstacle localization reliability. Indeed, networked control systems and
Internet of Things are key topics for the next generation of automated driving systems. Nevertheless,
this paper does not deal with networked control systems. This paper is focused on IoT LiDAR
sensors. Computational intelligence methods are then applied to determine the number of LiDAR
sensors that are required to increase sensor reliability for obstacle localization tasks. Further studies
related with the application of network-based fuzzy control and H control strategy to automated
driving systems will be explored in the future [17,18].
The paper consists of five sections. Following this introduction, the second section will show the
design and the implementation of several modules that employ self-tuning methods for improved
Sensors 2018, 18, 1508 3 of 16
reliability. Subsequently, a driving-assistance case study scenario for obstacle detection based on IoT
LiDAR model sensor information will be developed in Section 3. Additionally, the proposed
methodology will be validated with the minimum required number of sensors, to ensure LiDAR
sensor reliability in each scan. Finally, the conclusions and future research steps are presented.
2. Self-Tuning Method for Reliability in LiDAR Sensors Network
The self-tuning method for reliability in LiDAR sensor networks mainly consists of a computer-
aided system that enables efficient data interchange between the data transmitted from the IoT sensor
networks, managed by a control node network, and external modules that evaluate network
behavior. The supervisor node controller (SNC) is the component in charge of generating sensory
data through the simulation of sensor models; while the IoT assessment framework is the interface
with the external modules. Different simulation tools can be used for this purpose. On the one hand,
there is a 3D model simulator for automotive applications and on the other hand, an external
programming software with a set of toolkits to manage points cloud, clustering methods, pattern
recognition algorithms, and artificial intelligence (AI) based modelling strategies, among others.
2.1. Conceptual Design
The conceptual and architectural design of the proposed method is presented in this section.
Figure 1 shows the data interchange model between sensory nodes and actuation. The data
interchange component operates as data-sharing broker. The SNC and the IoT assessment framework
input information into the data interchange broker and collect information from it.
Figure 1. Conceptual design of self-tuning method. Iteration between IoT assessment framework and
supervisor node controller.
The SNC comprises different local control nodes containing IoT sensor network models,
distributed according to their functions. The distributed IoT sensors are in charge of capturing
sensory data and interchange these data with the SNC, in order to share it with other external
modules. It is important to highlight that data must necessarily pass through the supervisor, when
information is sent to and received from the different nodes of the IoT sensors network. However,
Sensors 2018, 18, 1508 4 of 16
the same procedure is not necessary when this data transfer is between the different IoT sensors that
comprise the sensor network.
The IoT assessment framework sends/receives data to/from the SNC. The first key component is
a module developed for a customized function directly linked to the local control nodes and the IoT
sensor network. The training procedure is performed by using classification methods based on
computational intelligence techniques such as k-nearest neighbors, multi-layer perceptron, support
vector machines, self-organizing map, Bayesian networks, etc. For example, on the one hand, a
module can be in charge of predicting the error in the localization of an object from data cloud points
given by a LiDAR sensor. On the other, it can involve a group of tasks for self-tuning (knowledge-
based learning algorithm). This module consists of a computational intelligence (CI) library of models
that contains various models with similar functions and a learning strategy (i.e., Q-learning) that
computes the actual threshold value in runtime, for the implementation of corrective actions. Both
methods can also be enriched at runtime with data received from the nodes of the IoT sensor network.
2.2. Implementation
The SNC, the local control nodes and IoT sensor network are designed and implemented using
a simulation tool for 3D models using Webots R2018a for automobiles [19]. In addition to its high
degree of potentiality when simulating sensors for driving assistance, Webots is able to interact with
other external software or programming languages, such as MATLAB, Python, Java, and Visual
#C/C++, among others. It should be noted that for modelling and simulation of sensors, any other
simulation tool for 3D sensor models can be selected.
An IoT assessment framework can therefore be implemented using any of the previously
mentioned software or external programming languages. One of these software programming
languages with an extensive set of libraries, MATLAB 2017b, was selected for developing the self-
tuning procedure. These tasks are carried out by two parallel execution threads, one of which is in
local mode with direct data transfer from the IoT sensor network, while the other functions at a global
level. The local thread (parallel execution 1) executes the current error prediction model from sensory
data provided by IoT sensor network. Subsequently, depending on the value of a certain threshold
that is calculated in runtime through a learning process, a set of corrective actions are performed.
This procedure is described in later sections.
On the other hand, the global thread (parallel execution 2) contains the CI model library with
other error prediction models and different performance indices. Later on, the library can also
enriched with the process simulation. During the simulation, new sensory data can be generated
providing new information on the environment in each interaction. Based on this continuous
information flow and the previous knowledge-based learning algorithm, the library executes a
parallel learning procedure for all error prediction models, to obtain a personalized setting for each
particular critical situation. Finally, once a new best configuration is generated, the corresponding
model in the IoT sensor network is then updated.
2.2.1. Supervisor Node Controller
The supervisor node controller manages the scenario in runtime and interchange data between
local control nodes or IoT sensors with other external modules. The overall operation of the 3D
scenario is managed by the SNC, in this case, Webots. Webots origins can be traced back to an
extension of robotic simulation software adapted to automobile simulations in a virtual environment.
A set of computational procedures manages the adaptation and transference of sensory information.
The data transfer is performed by means of different functionalities available in Webots. For example,
some available functions serve to create sensor models, such as LiDARs, stereo vision cameras, radar,
inertial, magnetic, gyroscopic, and GPS sensors that can be emulated with this software. In addition,
many obstacles and objects can be added to the scenario, such as simple vehicles, road segments,
traffic signals and lights, buildings, etc. A 3D traffic scenario can therefore be created, in order to
simulate the behavior of IoT sensor networks, which is incorporated in each local control node (i.e.,
a fully automated vehicle) for driving-assistance scenarios.
Sensors 2018, 18, 1508 5 of 16
2.2.2. Computational Intelligence Library for Modeling
The library of computational intelligence methods comprises three widely used techniques to
identify dynamic ecosystem patterns and for their predictive modeling. First, a multi-layer
perceptron (MLP) artificial neural network was selected. MLP is the most widely applied topology
as a universal approximator, guaranteeing higher performance with only one hidden layer and
therefore enabling good modeling capabilities [2022]. The MLP training process is based on the error
back-propagation algorithm. It is a supervised learning rule performed in two iterative steps. First,
the weights and biases are randomly initialized. Subsequently, the training samples are presented
one at a time. Then, for each s-th sample, the output of each neuron is computed from its inputs
( ) ( )
( ) [ ( )]
y s f u s=
which may be expressed as
( 1)
( ) ( ) ( ) ( )
( ) ( )
r r r r
j i j i j
u s w x s b
where, f () is the activation function; R is the total number of layers;
, the output of the j-th
neuron in the r-th layer for the s-th sample;
, the bias of this neuron;
, the weight connecting
the i-th neuron in the (r 1)-th layer with the j-th neuron in the r-th layer; and
is the input
from the i-th neuron in the (r 1)-th layer for the s-th sample.
The second selected technique was a k-nearest neighbors (kNN) clustering algorithm. kNN is one
of the simplest non-parametric classification methods with an easily interpretable output, low
calculation times, and high predictive power. Furthermore, it is widely used in classification and
regression problems for real applications [2326]. The algorithm is based on the predictions for a new
instance (x) by searching through the entire training set for the most similar instances of k (the
neighbors) and summarizing the output variable for those k instances. A Euclidean distance (ED) is
determined, to select the k instances in the training dataset that are the most similar to a new input.
( , ) ( )
i j ij
ED x x x x
The kNN algorithm is executed by following the five steps listed below:
Calculate the distance between test data and each row of training data
Sort the calculated distances in ascending order based on distance values
Select the top k rows from the sorted array
Select the most frequent class of these rows
Return the predicted class
Finally, a multiple linear regression technique was implemented to adjust the obstacle
localization model. The model can be represented as
, , ][,
Px b b x
y b b M
relating some input variables, x, with an output, y. Coefficients, b, are obtained by minimizing the
sum of the square of the difference between the predicted and the observed values,
, for
a set of m input-output pairs
Sensors 2018, 18, 1508 6 of 16
11 1 1
n nm n
x x y
x x y
 
 
 
 
 
the regression coefficients can be estimated through
)( TT
X Xb Xy
2.2.3. Threshold Detector and Q-Learning Procedure
The Q-learning algorithm is a model-free reinforcement learning technique. Specifically, Q-
learning can be used to find an optimal action-selection policy for any given (finite) Markov decision
process [27,28]. It learns an action-value function that ultimately yields the expected utility of taking
a given action in a given state and it then follows the optimal policy (off-policy learner). The algorithm
is based on a simple value iteration update. It assumes the old value and introduces a correction
based on the new information [29].
1 1 1 1
( , ) ( , ) ( max ( , ) ( , ))
t t t t t t t t t
Q s a Q s a r Q s a Q s a
+ + + +
+ + 
where, rt+1 is the reward observed after performing at in
, and α is the learning rate (0 < α ≤ 1).
The Q-learning algorithm is introduced in the closed-loop cycle (self-tuning), in order to
minimize the numbers of LiDAR sensors needed to guarantee good accuracy in the localization of a
detected obstacle within a minimum computation time, as illustrated in Figure 2. Table 1 shows
different rewards assigned for each detection threshold (γ).
Figure 2. Implementation of self-tuning procedure (knowledge-based learning algorithm).
Table 1. Q-learning reward matrix for detection threshold
Detection Threshold (γ) Ranges
Rewards for Number of LiDARs
This range of values is defined as a function of the obstacle prediction error calculated during
the classification step (see Figure 2). During the learning process, the algorithm recommends the
Sensors 2018, 18, 1508 7 of 16
optimal numbers of sensors to achieve a reliable obstacle detection, based on the previous knowledge
generated by the rewards obtained from similar situations learnt in the past. This self-turning
threshold produces a reduction of the computational cost and accelerates the prediction time that is
needed to detect an obstacle in the driving assistance environments.
3. IoT LiDAR Sensor Models for Obstacle DetectionA Case Study
A particular driving assistance scenario is defined, in order to evaluate the proposed self-tuning
methodology for its validation. In this use case, the methodology was applied to a LiDAR sensor
model to assess the reliability related to the error accuracy in the location of an object. In this section,
a model of error prediction from data of a single LiDAR sensor model was generated. Instead, this
same model will be used in a later section to evaluate and to establish the reliability of an IoT sensor
network. The method of generating a dataset for training and to validate the error prediction models
is described in the following sections (see Figure 3).
Figure 3. Overall scheme of the methodology of the self-tuning method.
3.1. Training Dataset from 3D Scenario Simulation
The Webots automobile simulation tool in conjunction with a four-layer LiDAR sensor model
was used, in order to generate a virtual driving traffic scenario. The scenario emulates the real test
track (a roundabout, traffic lights at a central crossroad and additional curves on an otherwise mainly
straight track) at the Centre for Automation and Robotics (CAR) in Ctra. Campo Real Km. 0.2,
Arganda del Rey (Madrid, Spain), which simulates an urban environment with pedestrians and a
fleet of six fully-automated vehicles in movement and a communications tower. A more detailed
description of this scenario and both the real and simulated parts can be found in [30]. Figure 4
illustrates the aerial view of some of these 3D scenarios in Webots Automobile for driving assistance.
A vehicle model (Toyota Prius), a camera image with objects that have been recognized and the
LiDAR point cloud are also illustrated in Figure 4. The objective of this camera is to obtain high-
precision object location for comparison with the location obtained by the LiDAR sensor, thereby
generating the error prediction models.
The fully sensorized vehicle model (Toyota Prius model) incorporated two on-board sensors,
one LiDAR sensor, and a 3D stereo vision camera (see Figure 4b). Both sensors were positioned inside
the vehicle, the LiDAR in the lower front section and the camera in the upper front section. Table 2
shows the specifications and localization of this sensor mounted on the vehicle.
Table 2. Specifications and location of both sensors on the vehicle
Ibeo Lux 4 Layers
Bumblebee 2 1394a
Bottom frontal
Front top
Horizontal field
120 deg. (35 to −50 deg.)
Size resolution max.
1034 × 776 pixels
Horizontal step
0.125 deg.
Pixel resolution
4.65 µm square pixels
Vertical field
3.2 deg.
Focal lengths
3.8 mm
Vertical step
0.8 deg.
Focal length/2.0
200 m
Horizontal Field of View
Update frequency
12.5 Hz
Frame rates
20 FPS
Sensors 2018, 18, 1508 8 of 16
Figure 4. (a) Aerial view of simulated 3D scenario in Webots automobile; (b) Vehicle model with
sensors incorporated; (c) Image captured by the camera and object detection procedure; (d) Point
cloud projection from the LiDAR sensor over objects once detected.
3.1.1. Benchmark Data
Following a simulation of 2 min and 56 s, data collection provided by the LiDAR sensor model
and a set of images captured by the 3D stereo vision camera were obtained. In total, a benchmark of
1031 scenes was available with the same number of LiDAR scans, captured images, and annotation
files with the localization of each recognized obstacle.
On the one hand, the camera benchmark contained 1031 images and their corresponding
annotation files, obtained from the camera-sensor object recognition algorithm, with the following
information: the localization (weight × height) of each object in the image (in pixels) and the size
(weight x height) of each object in the image. This sensor is a stereo vision camera with the following
specifications: 0.8 MP; resolution 1032 × 776, color, and 20 FPS.
On the other hand, the LiDAR benchmark contained 1031 scans. Each scan contained a 3-D point
cloud/scan. This small benchmark set was useful for exploring the accuracy of an obstacle in the
scene. Each scene contained the relative X, Y, Z position of an object in the first, the second, and the
third columns, showing the LiDAR localization coordinates at the scene (X0 = 0, Y0 = 0 and Z0 = 0),
and the fourth column listed the number of the corresponding layer.
In addition, the raw data from the LiDAR sensors required filtering and pre-processing in order
to facilitate determination of the location error of the object. Firstly, the points that map the road
asphalt at ground level and the vegetation were deleted. These points were mainly located at 20 cm
above ground level.
Secondly, fast indexing and search capabilities were required, in order to process the sensory
data. To do so, the point cloud was internally organized using a k-d tree structure; a space
partitioning data structure for mapping points onto a k-dimensional space [31]. The next data-
processing step consisted of extracting the points by mapping nearby obstacles that correspond to a
specific point-cloud sequence. A density-based spatial clustering of applications with a noise
(DBSCAN) algorithm was applied [32] for segmentation, which can segment the point cloud for each
available obstacle at the scene. The algorithm generated the points as clusters around each axis, and
Equation (7) was used to calculate the centroid of each segmented (X0, Y0, Z0) point cloud that
corresponded to each obstacle.
The final step was to compare each LiDAR calculated centroid with the actual location generated
by the object recognition algorithm of each obstacle, in order to obtain the accuracy error.
Sensors 2018, 18, 1508 9 of 16
Once the benchmark data set had been created, the next step was to generate the training data
set for the generation of the error prediction models. The spatial statistics of the point cloud were
then used as inputs for the error prediction model. Subsequently, the two errors that were taken into
account as outputs of the model were also explained.
3.1.2. Model Inputs
A group of spatial statistics were implemented, in order to standardize the model inputs
independently from the point cloud distribution [33]. The centrographic and directional distribution
of the point clouds were obtained from these spatial point pattern methods. Common centrographic
statistics for a point pattern are as follows: mean center, median center, standard deviational circle,
and standard deviational ellipse [34]. The mean center, MC, is characterized by geographic
coordinates {X, Y, Z} equal to the arithmetic means of the x-, y-, and z-coordinates of all the N points
in a pattern
1 1 1
( ) ; ( ) ; ( )
i i i
i i i
x y z
= = =
= = =
 
An alternative unique measure of the central spatial tendency of a point pattern that was used
is the center of minimum distance (often referred to as the median center), which is robust in the
presence of spatial outliers. Unlike the mean center, defining the median center, MedC, requires a
much more computationally complex iterative process to find a location that minimizes the Euclidean
distance, d, to all the points in a point pattern [35]
( ) ( ) ( )
2 2 2
1, 1
, , |min i N t A
t t t t t t
i i i
MedC x y z x x y y z z
+ + −
where, i defines each point in a point pattern, t is an iteration number, and {xt, yt, zt} is a location of
the median center of an iterative candidate. An important property of a point pattern is the degree of
its spatial spread. It can be characterized by the standard distance, SD, that is estimated as
( ) ( ) ( )
2 2 2
1 1 1
( ) ( ) ( )
n n n
i i i
i i i
x MC X x MC Y x MC Z
= = =
− −
= + +
 
where, xi, yi, and zi are the coordinates of point i{xi, yi, zi}, N is the total number of points, and MC(X),
MC(Y), and MC(Z) are the coordinates of the mean center. The standard distance that results from
the different average distances to a given centroid is usually graphically represented in a geographic
information system (GIS) environment by a standard deviational circle, centered on the mean center
with the radius equal to the standard distance.
Finally, the last spatial statistic used as inputs in this work is the third central moment (3rdCM)
[36]. This value represents the mean value of the cubic deviation in distance from each point to the
MC of each axis. The 3rdCM value is calculated as
( ) ( ) ( )
2 2 2
3rd ( ) ( ) ( )
= + + −
i i i
CM x MC X y MC Y z MC Z
where, xi, yi, and zi are the coordinates of point i{xi, yi, zi}, n is the total number of points and the MC
is the mean center by geographic coordinates {X, Y, Z}, as calculated in Equation (3).
3.1.3. Model Output
The outputs of these models are two figure s of merit of accuracy: the distance root mean
squared (DRMS) and the mean radial spherical error (MRSE). The first is a measure of data
tracked on the xy plane (2D) and the second is a measure of the data tracked in an xyz space (3D)
[37]. The DRMS and MRSE values were calculated as
Sensors 2018, 18, 1508 10 of 16
( ) ( )
 
i Actual ti i Actual ti
DRMS x x y y
= + −
( ) ( ) ( )
 
2 2 2
, , ,
i Actual ti i Actual ti i Actual ti
MRSE x x y y z z
= + + −
where, n is the number of readings for a dynamic tag during the time it is tracked, (xti, yti, zti) are the
coordinates of the tag at time ti, and (xActual,ti, yActual,ti, zActual,ti) are the actual coordinates of the tag at time ti.
3.2. Model Training and Initial Validation
The error-based prediction model library for the localization is defined, in order to estimate the
value of the figure of merits in terms of error (DRMS and MRSE), as a function of the parameters
extracted from the point cloud generated by the LiDAR sensors. Three models were considered in
this approach. First, a multilayer perceptron neural network with backpropagation (MLP)
comprising two hidden layers, with five neurons for each hidden layer, sigmoid activation functions,
and 104 epochs were trained. The learning rate (μ) initial value was 103 with a decrease factor ratio
of 101, an increase factor ratio of 10, and a maximum μ value of 1010. The minimum performance
gradient was 107. The adaptive value
is increased by 10 until the change above results in a reduced
performance value. The change is then made to the network and μ is decreased by 101. Training stops
when any of these conditions occurs as follows: the maximum number of epochs (repetitions) is
reached, or the maximum amount of time is exceeded, or the performance is minimized to the goal,
or the performance gradient falls below minimum gradient, or
exceeds 1010.
The second modeling technique was a two-neighbor k-nearest neighbors (kNN). Finally, a lineal
regression was also obtained by minimizing the sum of squared differences between the predicted
and the observed values.
A total of 1031 scans were extracted from the Webots simulator to generate the training and the
validation datasets. Subsequently, the scans were randomly divided into two datasets: 765 samples
for the training dataset (representing the 74% of the total of samples) and 255 samples to compose
the validation dataset (representing the other 26% of the total of all samples). The model correlation
coefficients (R2) were estimated for all the models implemented in the modeling library.
Table 3 showed the values obtained for each model based on the plane (DMRS) and the space
(MRSE) figures of merits described previously. As can be seen, the kNN algorithms represent, in both
cases, the best fitting parameters, with a 93% correlation between the xyz cloud of points
coordinates in the localization of each obstacle that was detected.
Finally, the DRMS tendency for all the models is shown in Figure 5, validating the best fit
obtained between the observed solution and the prediction values based on the kNN algorithm.
Nevertheless, in most cases, the MLP model presented a very similar behavior with the kNN model,
and not so very different from the linear regression model.
Table 3. Model correlation coefficients based on plane and space figures of merits
Correlation Coefficient (R2)
Linear Regression
Sensors 2018, 18, 1508 11 of 16
Figure 5. Prediction error behavior of the model library in the localization of obstacles by LiDAR point
4. Experimental Results
Additional experimental tests were conducted for the evaluation of the IoT on-chip LiDAR
sensor network performance and the self-tuning methodology performance in the dynamic obstacle
detention scenario. The aim was to define the best prediction model and the optimal number of
LiDAR sensors required to ensure the reliability of this sensor network. A set of critical conditions
were taken into account in the study.
In this case, the simulation time (43 s) for each traffic scenario was somewhat smaller, due to the
computational overload generated by the processing and storage of all the data provided by five
LiDARs plus two high-resolution cameras. In this experimental evaluation, the 3D stereo vision was
only activated when all the sensors, including the LiDAR sensors, were incapable of guaranteeing
sufficient reliability in terms of safety. In addition, the global scenario was the same as the one
described in Section 3. However, there are plenty of devices and objects in each scene with different
distributions. Some of these dynamic objects in the scenario are 4 buildings, 50 trees, 20 pedestrians,
10 small and medium vehicles, and 1 truck. Therefore, the main difference relies on the distribution
of the IoT sensory system in the fully automated vehicle (Toyota Prius) modelled in Webots (see
Figure 5).
Figure 6 represents the configuration of the IoT sensory system mounted in a fully automated
vehicle modeled in a Webots automobile.
Three equidistant LiDAR models (LiDAR 0/1/2) were attached at a low level on the front of the
vehicle, in order to expand the horizontal field of view. Two LiDAR devices (LiDAR 3/4) were placed
at a higher level on the front of the vehicle, on either side of the stereoscopic camera model, the
purpose of which is, in this case, to expand the vertical field of view. The aim of this IoT evaluation
is to demonstrate how an IoT sensory system incorporates a series of extended precision-related
measurement capabilities, with regard to the behavior of a single sensor with better specifications
than each isolated node of the sensory network. The setup of the IoT sensory system is summarized
in Table 4.
Table 4. Localization of each sensor that comprises the IoT sensory system
Localization (m)
3D Stereo Camera
Bumblebee 2
(0.0, 2.04, 1.2)
Ibeo Lux 4 layers
(0.0, 3.635, 0.5)
Ibeo Lux 4 layers
(−0.70, 3.64, 0.5)
Ibeo Lux 4 layers
(0.70, 3.64, 0.5)
Ibeo Lux 4 layers
(−0.55, 2.04, 1.2)
Ibeo Lux 4 layers
(0.55, 2.04, 1.2)
Sensors 2018, 18, 1508 12 of 16
Figure 6. Side (a); front (b); plan (c) and rear view (d) of on-board IoT sensory system setup in a
vehicle model.
The next step was to associate the same model of the error prediction model library to each
LiDAR sensor. The conceptual diagram of the self-tuning method is shown in Figure 7.
During simulations of each scenario, the information provided by each sensor was collected,
filtered and processed (as discussed in Section 3.1.1), the spatial statistics were calculated (as
described in Section 3.1.2) and applied as inputs to each model in the library of error prediction
models. These models estimate the accuracy of the localization for an object using the LiDAR sensors,
based on the DMRS and MRSE figures of merits. Table 5 lists the value of the correlation (R2) of each
type of model (ANN, kNN, and regression models) according to the number of LiDARs (1, 3, or 5)
used at each instant with the objective of expanding the field of view, both vertical and horizontal, of
the IoT LiDAR network in different critical situations. It must be stressed that the IoT sensors could
not exchange sensory information without a local pre-processing phase in each IoT node. In this case,
the shared information is the predicted error values obtained from each cloud point given by each
sensor rather than raw sensory information.
Figure 7. Flow diagram of the self-tuning method for the IoT sensor dynamic obstacle detection
Sensors 2018, 18, 1508 13 of 16
Table 5. Behavior of the correlation (R2) of each type of model according to the number of LiDAR
sensors used at any one given time
Model Correlation (R2)
From this table, the type of model within the library that showed a better correlation for outputs
of both the 2D and the 3D spatial error types, in comparison with the different IoT LiDAR system
configurations, could be extracted. With one LiDAR, the correlation values for both outputs were
similar to those obtained in Section 3.2.3. On the other hand, for a configuration of three sensors, the
value of this performance index improved notably in all models, highlighting kNN, as it is very close
to 100%. Although, in theory, an increased number of sensors will widen the field of view, it turned
out that the correlation with five sensors decreased in all models. One of the main causes is the
duplicity of information provided by too many sensors. This behavior is also reflected in Table 5
where accuracy can be seen to increase with the number of LiDAR sensors. However, the accuracy
also diminished when many LiDAR sensors brought too much sensory information. A problem could
be solved by using an optimized mesh distribution that avoids the duplication of space covered by
each LiDAR sensor.
Finally, during the simulation of this dynamic obstacle localization scenario, the learning and
self-tuning (knowledge-based learning algorithm) tasks were also validated to automatically set the
best prediction model and optimal number of LiDAR sensors needed to ensure reliability. The Q-
learning classification error matrix is shown in the Figure 8.
Figure 8. Q-learning classification error matrix.
As shown above, 67% of the scenarios can be appropriately addressed with a threshold between
01 (in other words, at a detection accuracy of over 99%) and only one LiDAR is needed. In total, 87%
of the scenarios can be solved using only one LiDAR, but if the threshold is larger than 10 (with less
than 90% of obstacle localization accuracy), then the use of three LiDARs is recommended,
considerably increasing the reliability of the multi-sensor-based system. Furthermore, it is worth
clarifying that only in 2% of cases are five LiDAR sensors needed, which is evidence of how the self-
tuning method can minimize the number of sensors required to achieve higher obstacle localization
reliability in the driving-assistance environments. In addition, the use of a minimal number of LiDAR
sensors that are simultaneously required for reliable and accurate prediction may be highlighted as
an important contribution. This rationalization of sensor use lowers energy consumption and saves
on computing resources. Therefore, the Q-learning method that processes the prediction model error
values at each instant can determine the minimum number of active LiDAR sensors that are needed,
Sensors 2018, 18, 1508 14 of 16
in order to ensure better reliability. It is only when this reliability cannot be assured with any
distribution of LiDAR sensors that the activation of the stereoscopic camera is enabled.
5. Conclusions
The design of a self-tuning method and its implementation for automatic selection of the number
of LiDAR sensors that are required to ensure object detection reliability in a IoT multisensory driving-
assistance scenario has been presented in this paper. Three simple techniques from the perspective
of industrial informatics have been considered in the implementation of the modeling library. A
linear regression method has demonstrated the direct correlation between the extracted point clouds
with obstacle localization. Secondly, the well-known multi-layer perceptron has once again been
confirmed as a suitable technique for modelling the main process characteristics; and a k-nearest
neighbors method has confirmed the suitability of clustering techniques to establish correlations
based on point dispersions. All the selected models have accurately reflected the behavior of the
selected variables and the statistical tests have confirmed the goodness-of-fit, highlighting that the
kNN algorithm obtains a correlation coefficient of over 90% in almost all the scenarios.
Additionally, a Q-learning strategy was also introduced, in order to minimize the number of
LiDAR sensors needed in each obstacle localization scenario and to ensure sensor reliability based on
global IoT sensor network information. The self-tuning procedure that has been explored for each
particular scenario has shown how many sensors are required for accurate obstacle localization in
each scan. Based on those results, the proposed method has fulfilled two main criteria: the best model-
based fitting and self-tuning management of the computational resources (smaller number of LiDAR
real required for each particular situation) necessary for improvements in obstacle localization
reliability on IoT LiDAR sensor networks. The accuracy and generalization of the proposed method,
developed in a virtual driving traffic scenario with a Webots Automobile simulation tool, has solved
67% of the scenarios with one LiDAR with an obstacle localization accuracy of over 99%. Finally, the
embedding of the proposed self-tuning methodology for validation in real driving environments is
foreseen in future contributions to the European IoSENSE project (
Author Contributions: R.E.H. reviewed all technical and scientific aspects of the article. A.V. was in charge of
the implementation of both the library models and the reinforcement learning algorithm. F.C. and G.B. designed
and implemented the scenario and the self-tuning methodology, and drafted the paper.
Acknowledgments: The authors wish to thank the support provided through the IoSENSE project: Flexible
FE/BE Sensor Pilot Line for the Internet of Everything, funded by the Electronic Component Systems for
European Leadership Joint (ECSEL) Undertaken under grant agreement No. 692480.
Conflicts of Interest: The authors declare no conflict of interest.
1. Pajares Redondo, J.; Prieto González, L.; García Guzman, J.; Boada, B.L.; Díaz, V. Vehiot: Design and
evaluation of an iot architecture based on low-cost devices to be embedded in production vehicles. Sensors
2018, 18, 486.
2. Krasniqi, X.; Hajrizi, E. Use of iot technology to drive the automotive industry from connected to full
autonomous vehicles. IFAC-PapersOnLine 2016, 49, 269274.
3. Vivacqua, R.; Vassallo, R.; Martins, F. A low cost sensors approach for accurate vehicle localization and
autonomous driving application. Sensors 2017, 17, 2359.
4. Kempf, J.; Arkko, J.; Beheshti, N.; Yedavalli, K. Thoughts on reliability in the internet of things. In
Interconnecting Smart Objects with the Internet Workshop; Internet Architecture Board: Boston, MA, USA, 2011;
Volume 1, pp. 14.
5. Ahmad, M. Reliability models for the internet of things: A paradigm shift. In Proceedings of the 2014 IEEE
International Symposium on Software Reliability Engineering Workshops, Naples, Italy, 36 November
2014; pp. 5259.
6. Xiao, L.; Wang, R.; Dai, B.; Fang, Y.; Liu, D.; Wu, T. Hybrid conditional random field based camera-LiDAR
fusion for road detection. Inf. Sci. 2018, 432, 543558.
Sensors 2018, 18, 1508 15 of 17
7. Zeng, Y.; Yu, H.; Dai, H.; Song, S.; Lin, M.; Sun, B.; Jiang, W.; Meng, M. An improved calibration method
for a rotating 2D LiDAR system. Sensors 2018, 18, 497.
8. Bein, D.; Jolly, V.; Kumar, B.; Latifi, S. Reliability modeling in wireless sensor networks. Int. J. Inf. Technol.
2005, 11, 18.
9. AboElFotoh, H.M.F.; Iyengar, S.S.; Chakrabarty, K. Computing reliability and message delay for
cooperative wireless distributed sensor networks subject to random failures. IEEE Trans. Reliab. 2005, 54,
10. Hu, S.; Li, Z.; Zhang, Z.; He, D.; Wimmer, M. Efficient tree modeling from airborne LiDAR point clouds.
Comput. Graph. 2017, 67, 113.
11. Castaño, F.; Beruvides, G.; Haber, R.; Artuñedo, A. Obstacle recognition based on machine learning for on-
chip LiDAR sensors in a cyber-physical system. Sensors 2017, 17, 2109.
12. Shi, B.; Han, L.; Yan, H. Adaptive clustering algorithm based on knn and density. Pattern Recognit. Lett.
2018, 104, 3744.
13. Zhang, S.; Cheng, D.; Deng, Z.; Zong, M.; Deng, X. A novel knn algorithm with data-driven k parameter
computation. Pattern Recognit. Lett. 2017, doi:10.1016/j.patrec.2017.09.036.
14. Połap, D.; Kęsik, K.; Książek, K.; Woźniak, M. Obstacle detection as a safety alert in augmented reality
models by the use of deep learning techniques. Sensors 2017, 17, 2803.
15. Stone, P. Q-learning. In Encyclopedia of Machine Learning and Data Mining; Sammut, C., Webb, G.I., Eds.;
Springer US: Boston, MA, USA, 2017; pp. 10331033.
16. Haber, R.E.; Juanes, C.; del Toro, R.; Beruvides, G. Artificial cognitive control with self-x capabilities: A case
study of a micro-manufacturing process. Comput. Ind. 2015, 74, 135150.
17. Wang, T.; Qiu, J.; Gao, H.; Wang, C. Network-based fuzzy control for nonlinear industrial processes with
predictive compensation strategy. IEEE Trans. Syst. Man Cybern. Syst. 2017, 47, 21372147.
18. Tian, E.; Yue, D. Decentralized fuzzy H filtering for networked interconnected systems under
communication constraints. Neurocomputing 2016, 185, 2836.
19. Michel, O. Cyberbotics Ltd. Webots™: Professional mobile robot simulation. Int. J. Adv. Robot. Syst. 2004,
1, 5.
20. Fe, I.L.; Beruvides, G.; Quiza, R.; Haber, R.; Rivas, M. Automatic selection of optimal parameters based on
simple soft computing methods. A case study on micro-milling processes. IEEE Trans. Ind. Inform. 2018,
21. Alique, A.; Haber, R.E.; Haber, R.H.; Ros, S.; Gonzalez, C. Neural network-based model for the prediction
of cutting force in milling process. A progress study on a real case. In Proceedings of the IEEE International
Symposium on Intelligent Control, Patras, Greece, 1719 June 2000; pp. 121125.
22. Haber, R.E.; Alique, J.R. Nonlinear internal model control using neural networks: An application for
machining processes. Neural Comput. Appl. 2004, 13, 4755.
23. Aziz, S.; Mohamed, E.A.; Youssef, F. Traffic sign recognition based on multi-feature fusion and elm
classifier. Procedia Comput. Sci. 2018, 127, 146153.
24. Khaldi, B.; Harrou, F.; Cherif, F.; Sun, Y. Self-organization in aggregating robot swarms: A dw-knn
topological approach. Biosystems 2018, 165, 106121.
25. Beruvides, G.; Castaño, F.; Haber, R.E.; Quiza, R.; Villalonga, A. Coping with complexity when predicting
surface roughness in milling processes: Hybrid incremental model with optimal parametrization.
Complexity 2017, 2017, doi:10.1155/2017/7317254.
26. Penedo, F.; Haber, R.E.; Gajate, A.; Del Toro, R.M. Hybrid incremental modeling based on least squares
and fuzzy k-nn for monitoring tool wear in turning processes. IEEE Trans. Ind. Inform. 2012, 8, 811818.
27. Seel, N.M. Greedy q-learning. In Encyclopedia of the Sciences of Learning; Springer US: Boston, MA, USA,
2012; pp. 13881388.
28. Beruvides, G.; Juanes, C.; Castaño, F.; Haber, R.E. A self-learning strategy for artificial cognitive control
systems. In Proceedings of the 2015 IEEE International Conference on Industrial Informatics, Búzios, Brazil,
35 June 2015; pp. 11801185.
29. Chincoli, M.; Liotta, A. Self-learning power control in wireless sensor networks. Sensors 2018, 18, 375.
30. Artuñedo, A.; del Toro, R.; Haber, R. Consensus-based cooperative control based on pollution sensing and
traffic information for urban traffic networks. Sensors 2017, 17, 953.
31. Zhou, K.; Hou, Q.; Wang, R.; Guo, B. Real-time kd-tree construction on graphics hardware. ACM Trans.
Graph. 2008, 27, 111.
Sensors 2018, 18, 1508 16 of 16
32. Ester, M.; Kriegel, H.-P.; Sander, J.; Xu, X. A density-based algorithm for discovering clusters in large spatial
databases with noise. In Proceedings of the Second International Conference on Knowledge Discovery and
Data Mining, Portland, Oregon, 24 August 1996; AAAI Press: Portland, Oregon, 1996; pp. 226231.
33. Lisitsin, V. Spatial data analysis of mineral deposit point patterns: Applications to exploration targeting.
Ore Geol. Rev. 2015, 71, 861881.
34. De Smith, M.J.; Goodchild, M.F.; Longley, P. Geospatial Analysis: A Comprehensive Guide to Principles,
Techniques and Software Tools; Troubador Publishing Ltd.: Leicester, UK, 2007.
35. Burt, J.E.; Barber, G.M.; Rigby, D.L. Elementary Statistics for Geographers; Guilford Press: New York, NY,
USA, 2009.
36. Premebida, C.; Ludwig, O.; Nunes, U. Exploiting LiDAR-based features on pedestrian detection in urban
scenarios. In Proceedings of the 2009 12th International IEEE Conference on Intelligent Transportation
Systems, St. Louis, MO, USA, 47 October 2009; pp. 16.
37. Maalek, R.; Sadeghpour, F. Accuracy assessment of ultra-wide band technology in locating dynamic
resources in indoor scenarios. Autom. Constr. 2016, 63, 1226.
© 2018 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access
article distributed under the terms and conditions of the Creative Commons Attribution
(CC BY) license (
... A trailing vehicle is able to follow the leader's trajectory using a combination of fuzzy logic controllers. Castaño et al. presented a self-tuning method to maximize the reliability of LiDAR sensors network for obstacle detection in the Internet of Things mobility scenarios [9]. The experimental results demonstrated that the selftuning method is an appropriate strategy to increase the reliability of the sensor network while minimizing detection thresholds. ...
... Then the candidate routes that exceed the preset deadline to reach the destination are removed from the list in accordance with the list of constraints given in Eqs. (2)(3)(4)(5)(6)(7)(8)(9). Next, we derive the value inside the Min (⋅) operator in Eq. (1) for each qualified candidate route by computing the inner product of the three weights and the corresponding objectives. ...
... The definitions of the rest of the parameters are identical to those used in Eqs.(2)(3)(4)(5)(6)(7)(8)(9). ...
Full-text available
Traffic congestion in urban areas has become a critical problem that municipal governments cannot overlook. Meanwhile, mixed traffic systems containing both autonomous and human-driven electric vehicles ramp up the challenge for traffic management in urban areas. Although numerous researchers have proposed traffic control heuristics to alleviate traffic congestion problems in the recent literature, scant research has addressed the joint problems of route and charging strategies for electric vehicles along with urban traffic congestion prevention. Accordingly, this work tackles the complex task of traffic management in urban areas during peak periods by using practical congestion prevention strategies that consider the characteristics of mixed traffic flows and the charging demands of electric vehicle users. Notably, we apply support vector regressions to compute the charging time at each charging point and the traverse time of an electric vehicle at each road segment/intersection, based on historical traffic data. The simulation results reveal that the proposed algorithms are feasible because they can avoid possible occurrences of traffic congestion during rush hours and provide the routes and charging options that are chosen by electric vehicle users.
... Posiblemente se hará uso de sistemas de red veloces (Diario Siglo XXI, 2019) por satélite u onda corta, conectadas permanente o semipermanentemente(Fig. 6), formando un sistema de IA por canales de información con funciones 14 Posiblemente, y de omitir los deberes no-dilemáticos, tales como: pagar los impuestos proporcionales respectivos, atender a heridos en un accidente en el que se puede ayudar y nadie más puede, llamar a emergencias ante ciertas alertas percibidas y no contestadas… o llevar mascarilla y vacunarse ante una pandemia mundial; las situaciones morales más extremas, de dilema, que se pueden vivir hoy en Occidente suelen ser a lo sumo intentar obrar con bastante riesgo propio en un posible crimen, ataque terrorista o catástrofe para ayudar a otros, intervenir o no en defensa de alguien en sus derechos ante una agresión, etc.La ética de la automatización en vehículos y enfoques alternos para problemas morales actuales Óscar Frederic Teixidó Durán -Rev Bio y Der. 2023; 57: 153-180 DOI 10.1344/rbd2023.57.38024 -ISSN 1886-5887 | 174(representables en grafos), logrando datos radar con tecnología LiDAR -Laser Imaging Detection and Ranging-(Castaño et al., 2018). ...
Full-text available
En el presente trabajo se utiliza el problema ético de la automatización en transportes y vehículos dotados de inteligencia artificial (IA) para un análisis crítico de las diferentes posiciones éticas y morales clásicas. Debido a la intersección de la IA con los campos de investigación de la bioética, la nomoética y la tecnoética, existe un hilo conductor en sus fundamentos. Además, se juzga necesario para este propósito, así como por rigurosidad, hacer uso de una cantidad mínima de herramientas formales y exponer un proyecto de una teoría ética actualizada, sistémica y materialista, que sobrepase a sus sistemas rivales. Así, se pretende ofrecer las bases para una teoría alterna más fecunda y explicar cómo las posiciones éticas tradicionales pueden ser obsoletas, si no directamente inaplicables, para afrontar la automatización inteligente y problemas morales similares, en la actualidad.
... In recent years, there has been an emphasis on LiDAR-based obstacle detection methods (Onda et al., 2018;Asvadi et al., 2016;Peng et al., 2015) due to the high resolution, anti-interference capability, and spatial measurability of LiDAR data. Such methods include negative obstacle detector (NODR)-based and support vector machine (SVM) algorithms (Larson and Trivedi, 2011)), the self-tuning methodology (Castaño et al., 2018) and the clustering algorithm based on relative distance and density (CBRDD) (Zheng et al., 2019). to enable obstacle detection. ...
Full-text available
An obstacle element map is the basis of indoor navigation space subdivision. Effective and reasonable indoor space subdivision can improve the fineness of indoor navigation path planning. To divide an indoor navigation space more finely, an indoor obstacle element map construction method based on scene-aware a priori rules for obstacles is proposed. This method divides indoor space into navigable space (N-space), semi-navigable space (SN-space), and non-navigable space (NN-space) based on the flexible space subdivision framework (FSS) and classifies indoor obstacles into four categories, walls, doors, movable obstacles (MOs), and immovable obstacles (IMOs). First, by analyzing the layout and mobility of common indoor elements, the obstacle prior rules for seven types of indoor scenes (including hospitals, residences, classrooms, office buildings, transportation hubs, restaurants, and shopping malls) are formulated, and a method for obtaining obstacle prior rules based on scene recognition is proposed. Then, obstacles and their mobility are extracted from the semantic segmentation results of point clouds based on the acquired rules and mobility indicators. Finally, the extracted obstacles are processed to construct the obstacle element map, wall and door elements are fitted by the random sample consensus (RANSAC) algorithm, and IMOs and MOs are processed by Euclidean clustering. The experimental results show that the proposed method can effectively determine four types of indoor obstacles, and the overall accuracy in simple scenes is 85.22 %, while in complex scenes it is 97.96 %. Under the experimental environment (CPU: Intel Core i7-7700K; Memory: 16 GB and 2400 MHz; OS: Win10) in this paper, the construction of a simple scene obstacle element map can be completed in 5 s. In addition to the spatial information of the indoor elements, the obstacle element map also contains information about whether the elements can be moved, which provides the basis for the subdivision of the indoor space.
... They have the advantages of reducing traffic accidents, improving traffic efficiency, reducing fuel costs, and protecting the environment [1,2]. Under unexpected conditions such as the failure of the sensory equipment of AVs and the sudden entry of objects in front of them [3], AVs will encounter situations where collision is unavoidable, i.e., a moral dilemma. ...
Full-text available
How decisions are made when autonomous vehicles (AVs) are faced with moral dilemmas is still a challenge. For this problem, this paper proposed the concept of common principles, which were drawn from the general public choice and could be generally accepted by society. This study established five moral dilemma scenarios with variables including the number of sacrifices, passenger status, presence of children, decision-making power subjects, and laws. Based on existing questionnaire data, we used gray correlation analysis to analyze the influence of the individual and psychological factors of participants in decision-making. Then, an independent sample t-test and analysis of covariance were selected to analyze the influence relationship between individual and psychological factors. Finally, by induction statistics of decision choices and related parameters of participants, we obtain common principles of autonomous vehicles, including the principle of protecting law-abiding people, the principle of protecting the majority, and the principle of protecting children. The principles have different priorities in different scenarios and can meet the complex changes in moral dilemmas. This study can alleviate the contradiction between utilitarianism and deontology, the conflict between public needs and individualized needs, and it can provide a code of conduct for ethical decision-making in future autonomous vehicles.
... Internet of Things e online monitoring of important operation parameters in various links of the ubiquitous power Internet of ings can strengthen the prediction, prevention, and control of equipment status from the aspects of safety, reliability, adjustability, and antidisturbance and provide intelligent auxiliary decision-making for reliable monitoring of transmission lines and scientific operation and maintenance of distribution links [22,23], thus further improving the operation level of power grid, strengthening the two-way interaction with users, and expanding new value-added services [24]. e realization of the above objectives relies on thorough the information perception, reliable data transmission, efficient network construction, and intelligent management and data processing analysis technology of massive sensing information [25]. ...
Full-text available
With the rapid development of the economy and society, the low efficiency and high loss of the traditional power grid can no longer meet the growing social demand, and the power grid market is facing a reform. Smart grid, as a next-generation power system, it can effectively improve the performance of traditional power grids. The ubiquitous power Internet of Things (UPIOT) replaces the traditional grids with efficient, safe, reliable, and flexible new grids, improves the utilization efficiency of the grid, reduces the loss of the power grid in the transmission process, and can meet the needs of different types of markets and users. As an advanced information acquisition and processing technology, wireless sensor networks have been widely used in medical, industrial, agricultural, commercial, and public management fields. It is an important means to promote future economic development and build a harmonious society. In the power system, wireless sensor network technology can be widely used in many fields such as line fault location, real-time monitoring, smart meter reading, and relay protection. In this paper, the basic concepts and overall architecture of ubiquitous power Internet of Things are summarized. Then, we summarize the research status of the wireless sensor network in smart grid, including power equipment, line monitoring, smart grid wireless automatic meter reading, distribution network relay protection, power assets life-cycle management, power grid fault location, and power grid fault diagnosis. In view of the technical characteristics of wireless sensor networks, combined with the production links of power systems, the application framework of wireless sensor network technology in the power systems is constructed. The application of wireless sensor networks is prospected from the aspects of network development of relay protection, application research of smart substation, application research of power grid catastrophe, security protection of power system, and deep-seated ubiquitous power Internet of Things.
... However, since determining the last vehicle is difficult, this approach is limited to a local solution. Castaño et al. (2017) and Castaño et al. (2018) introduced the obstacle detection method where the object is recognized using Support Vector Machine. Collision avoidance is accomplished using a Multi-Layer Perceptron and a Self-Organizing Map. ...
Full-text available
The traffic congestion and the rise in the number of vehicles have become a grievous issue, and it is focused worldwide. One of the issues with traffic management is that the traffic light’s timer is not dynamic. As a result, one has to remain longer even if there are no or fewer vehicles, on a roadway, causing unnecessary waiting time, fuel consumption and leads to pollution. Prior work on smart traffic management systems repurposes the use of Internet of things, Time Series Forecasting, and Digital Image Processing. Computer Vision-based smart traffic management is an emerging area of research. Therefore a real-time traffic light optimization algorithm that uses Machine Learning and Deep Learning Techniques to predict the optimal time required by the vehicles to clear the lane is presented. This article concentrates on a two-step approach. The first step is to obtain the count of the independent category of the class of vehicles. For this, the You Only Look Once version 4 (YOLOv4) object detection technique is employed. In the second step, an ensemble technique named eXtreme Gradient Boosting (XGBoost) for predicting the optimal time of the green light window is implemented. Furthermore, the different implemented versions of YOLO and different prediction algorithms are compared with the proposed approach. The experimental analysis signifies that YOLOv4 with the XGBoost algorithm produces the most precise outcomes with a balance of accuracy and inference time. The proposed approach elegantly reduces an average of 32.3% of waiting time with usual traffic on the road.
In recent years, there has been a drastic increase in research related to autonomous driving applications, getting autonomous vehicles to circulate and engage in vehicular traffic. However, there are still difficulties in achieving reliable autonomous driving. This is due to the unexpected actions of the entities involved in the autonomous vehicle’s journey, the lack of interaction of the vehicle with the environment, the reliability of the sensors in capturing the environment, among others. These problems have already caused several accidents involving autonomous vehicles. In order to overcome these drawbacks, researchers rely on 3D simulation software with modules dedicated to autonomous driving applications with the aim of shortening research times and especially the validation of dangerous scenarios. The research in this article proposes the simulation of a vehicle that performs automatic guidance while avoiding obstacles along its path. The proposal is to evaluate the reliability of the angle and distance of an obstacle detected by a LiDAR sensor. These two variables will also be obtained and compared by means of a geometric model that involves the additional implementation of a GPS sensor and a Compass sensor over the LiDAR sensor. Automatic vehicle guidance is achieved through the implementation of a Camera sensor that captures images that are processed to track the yellow line of the road. The results obtained can be used to plan methods and techniques for more accurate and reliable maps.
This research paper is focused in analyzes the behavior and performance of a control law that has been validated in a previous study in a three dimensions (3D) simulation software, now evaluated in a Mazda 6 vehicle with the objective that it travels a previously designed trajectory. This is possible, thanks to the use of a Global Positioning System (GPS) sensor as well as additional sensors that allow us to know the direct kinematics of the vehicle at each instant of time. The experimental tests were carried out with the use of a Raspberry pi 3 b + (small single-board computers) that is in charge of receiving the information from the sensors and then processed this information so that the control law allows the control actions sent to the actuators. Finally, results are presented to be considered for future research that allows the implementation of a fully autonomous driving system.KeywordsKinematic controlGPS sensorAutonomous driving
Full-text available
Reinforcement learning (RL) methods can successfully solve complex optimization problems. Our article gives a systematic overview of major types of RL methods, their applications at the field of Industry 4.0 solutions, and it provides methodological guidelines to determine the right approach that can be fitted better to the different problems, and moreover, it can be a point of reference for R&D projects and further researches.
Full-text available
The intersection line information of the point cloud between the coal wall and the roof can not only accurately reflect the direction information of the scraper conveyor but also provide a preliminary basis for realizing the intelligent coal mine. However, the indirect method of using deep learning to segment the point cloud of coal mine working face cannot make full use of the rich information provided by the point cloud data. The direct method of using deep learning to segment the point cloud ignores the local feature relationship between points. Therefore, we propose to use dynamic graph convolution neural networks (DGCNNs) to segment the point cloud of the coal wall and roof so as to obtain the intersection line between them. First, in view of the characteristics of heavy dust and strong electromagnetic interference in the environment of the coal mine working face, we have installed an underground inspection robot so that we use light detection and ranging to obtain the point cloud of the coal mine working face. At the same time, we put forward a fast labeling method of the point cloud of the coal mine working face and an efficient training method of the depth neural network. Second, on the basis of edge convolution, being the greatest innovation of DGCNNs, we analyze the influence of the number of layers, K value, and output feature dimension of edge convolution on the effect of DGCNNs segmenting the point cloud of the coal mine working face and obtaining the intersection line of the coal wall and roof. Finally, we compare DGCNNs with PointNet and PointNet++. The results show that the DGCNN exhibits the best performance. What is more, the results provide a research foundation for the application of DGCNNs in the field of energy. Last but not least, the research results provide a direct and key basis for the adjustment of the scraper conveyor, which is of great significance for an intelligent coal mine working face and accurate construction of a geological information model.
Full-text available
This paper proposes a novel and efficient method for traffic sign recognition based on combination of complementary and discriminative feature sets. The extracted features are the histogram of oriented gradients (HOG) feature, Gabor feature and Compound local binary pattern (CLBP) feature. The classification is performed using the extreme learning machine (ELM) algorithm. Performances of the proposed approach are evaluated on both German Traffic Sign Recognition Benchmark (GTSRB) and Belgium Traffic Sign Classification (BTSC) Datasets respectively. The results of the experimental work demonstrate that each feature yields fairly high accuracy and the combination of three features has shown good complementariness and yielded fast recognition rate and is more adequate for real-time application as well.
Full-text available
This paper presents an improved calibration method of a rotating two-dimensional light detection and ranging (R2D-LIDAR) system, which can obtain the 3D scanning map of the surroundings. The proposed R2D-LIDAR system, composed of a 2D LIDAR and a rotating unit, is pervasively used in the field of robotics owing to its low cost and dense scanning data. Nevertheless, the R2D-LIDAR system must be calibrated before building the geometric model because there are assembled deviation and abrasion between the 2D LIDAR and the rotating unit. Hence, the calibration procedures should contain both the adjustment between the two devices and the bias of 2D LIDAR itself. The main purpose of this work is to resolve the 2D LIDAR bias issue with a flat plane based on the Levenberg-Marquardt (LM) algorithm. Experimental results for the calibration of the R2D-LIDAR system prove the reliability of this strategy to accurately estimate sensor offsets with the error range from -15 mm to 15 mm for the performance of capturing scans.
Full-text available
Nowadays, the current vehicles are incorporating control systems in order to improve their stability and handling. These control systems need to know the vehicle dynamics through the variables (lateral acceleration, roll rate, roll angle, sideslip angle, etc.) that are obtained or estimated from sensors. For this goal, it is necessary to mount on vehicles not only low-cost sensors, but also low-cost embedded systems, which allow acquiring data from sensors and executing the developed algorithms to estimate and to control with novel higher speed computing. All these devices have to be integrated in an adequate architecture with enough performance in terms of accuracy, reliability and processing time. In this article, an architecture to carry out the estimation and control of vehicle dynamics has been developed. This architecture was designed considering the basic principles of IoT and integrates low-cost sensors and embedded hardware for orchestrating the experiments. A comparison of two different low-cost systems in terms of accuracy, acquisition time and reliability has been done. Both devices have been compared with the VBOX device from Racelogic, which has been used as the ground truth. The comparison has been made from tests carried out in a real vehicle. The lateral acceleration and roll rate have been analyzed in order to quantify the error of these devices.
Full-text available
Current trends in interconnecting myriad smart objects to monetize on Internet of Things applications have led to high-density communications in wireless sensor networks. This aggravates the already over-congested unlicensed radio bands, calling for new mechanisms to improve spectrum management and energy efficiency, such as transmission power control. Existing protocols are based on simplistic heuristics that often approach interference problems (i.e., packet loss, delay and energy waste) by increasing power, leading to detrimental results. The scope of this work is to investigate how machine learning may be used to bring wireless nodes to the lowest possible transmission power level and, in turn, to respect the quality requirements of the overall network. Lowering transmission power has benefits in terms of both energy consumption and interference. We propose a protocol of transmission power control through a reinforcement learning process that we have set in a multi-agent system. The agents are independent learners using the same exploration strategy and reward structure, leading to an overall cooperative network. The simulation results show that the system converges to an equilibrium where each node transmits at the minimum power while respecting high packet reception ratio constraints. Consequently, the system benefits from low energy consumption and packet delay.
Full-text available
The complexity of machining processes relies on the inherent physical mechanisms governing these processes including nonlinear, emergent, and time-variant behavior. The measurement of surface roughness is a critical step done offline by expensive quality control procedures. The surface roughness prediction using an online efficient computational method is a difficult task due to the complexity of machining processes. The paradigm of hybrid incremental modeling makes it possible to address the complexity and nonlinear behavior of machining processes. Parametrization of models is, however, one bottleneck for full deployment of solutions, and the optimal setting of model parameters becomes an essential task. This paper presents a method based on simulated annealing for optimal parameters tuning of the hybrid incremental model. The hybrid incremental modeling plus simulated annealing is applied for predicting the surface roughness in milling processes. Two comparative studies to assess the accuracy and overall quality of the proposed strategy are carried out. The first comparative demonstrates that the proposed strategy is more accurate than theoretical, energy-based, and Taguchi models for predicting surface roughness. The second study also corroborates that hybrid incremental model plus simulated annealing is better than a Bayesian network and a multilayer perceptron for correctly predicting the surface roughness.
Full-text available
Augmented reality (AR) is becoming increasingly popular due to its numerous applications. This is especially evident in games, medicine, education, and other areas that support our everyday activities. Moreover, this kind of computer system not only improves our vision and our perception of the world that surrounds us, but also adds additional elements, modifies existing ones, and gives additional guidance. In this article, we focus on interpreting a reality-based real-time environment evaluation for informing the user about impending obstacles. The proposed solution is based on a hybrid architecture that is capable of estimating as much incoming information as possible. The proposed solution has been tested and discussedwith respect to the advantages and disadvantages of different possibilities using this type of vision.
Full-text available
Autonomous driving in public roads requires precise localization within the range of few centimeters. Even the best current precise localization system based on the Global Navigation Satellite System (GNSS) can not always reach this level of precision, especially in an urban environment, where the signal is disturbed by surrounding buildings and artifacts. Laser range finder and stereo vision have been successfully used for obstacle detection, mapping and localization to solve the autonomous driving problem. Unfortunately, Light Detection and Ranging (LIDARs) are very expensive sensors and stereo vision requires powerful dedicated hardware to process the cameras information. In this context, this article presents a low-cost architecture of sensors and data fusion algorithm capable of autonomous driving in narrow two-way roads. Our approach exploits a combination of a short-range visual lane marking detector and a dead reckoning system to build a long and precise perception of the lane markings in the vehicle’s backwards. This information is used to localize the vehicle in a map, that also contains the reference trajectory for autonomous driving. Experimental results show the successful application of the proposed system on a real autonomous driving situation.
Nowadays, the application of novel soft-computing methods to new industrial processes is often limited by the actual capacity of the industry to assimilate state-of-the-art computational methods. The selection of optimal parameters for efficient operation is challenging in micro-scale manufacturing processes because of intrinsic nonlinear behaviors and reduced dimensions. In this article, a decision-making system for selecting optimal parameters in micro-milling operations is designed and implemented using simple and, efficient soft-computing techniques. The procedure primarily consists of four steps: an experimental characterization; the modelling by means of a multilayer perceptron; the two-objective optimization using the cross-entropy method and a decision-making procedure using a fuzzy inference system. In order to evaluate the proposed system, micro-milling processes of titanium-based and a tungsten-copper alloys are considered. The experimental study demonstrates the effectiveness of the proposed solution for automatically decision making based on simple soft-computing methods and its successfully application to a truly industrial challenge.
Although many clustering algorithms have been proposed, they all have various limitations. Existing clustering algorithms usually require the user to set the appropriate threshold parameters and those parameters are usually not adaptive. In this paper, a new clustering method is proposed. Firstly, the k nearest neighbors of all samples is calculated, and then a density method based on kNN is used to complete the clustering process. In order to achieve this goal, a statistics alpha is proposed for measuring the degree of denseness or sparseness of distribution of the entire samples, and a statistical model is proposed to calculate the local parameters, beta and gamma. That is, the category of each sample is determined by the combination of kNN with density. That is to say, the total number of samples determines an appropriate k value to obtain the ordered kNN of every sample, and then the radius of surrounding region (SR) is calculated by 5NN of all samples. Furthermore, the density of every sample is calculated. In the whole process of clustering, the global threshold is determined by the density distribution of all samples, and then the local threshold is self-adaptive. All sample density is sorted to search automatically for clusters from the highest point of density of the distribution of all samples. The algorithm can not only discover clusters of arbitrary shapes and automatically remove noise and outliers, but also it can find clusters with different densities and those with internal density variation. The results of experiments on the synthetic data and human face image data demonstrate that the algorithm is effective. The code is available under the following URL:
In certain swarm applications, where the inter-agent distance is not the only factor in the collective behaviours of the swarm, additional properties such as density could have a crucial effect. In this paper, we propose applying a Distance-Weighted K-Nearest Neighbouring (DW-KNN) topology to the behaviour of robot swarms performing self-organized aggregation, in combination with a virtual physics approach to keep the robots together. A distance-weighted function based on a Smoothed Particle Hydrodynamic (SPH) interpolation approach, which is used to evaluate the robot density in the swarm, is applied as the key factor for identifying the K-nearest neighbours taken into account when aggregating the robots. The intra virtual physical connectivity among these neighbours is achieved using a virtual viscoelastic-based proximity model. With the ARGoS based-simulator, we model and evaluate the proposed approach, showing various self-organized aggregations performed by a swarm of N foot-bot robots. Also, we compared the aggregation quality of DW-KNN aggregation approach to that of the conventional KNN approach and found better performance.