Professional Documents
Culture Documents
Sensor-Fusion_and_Tracking_Method_for_Indoor_Vehicles_With_Low-Density_UHF-RFID_Tags
Sensor-Fusion_and_Tracking_Method_for_Indoor_Vehicles_With_Low-Density_UHF-RFID_Tags
Sensor-Fusion_and_Tracking_Method_for_Indoor_Vehicles_With_Low-Density_UHF-RFID_Tags
Abstract— This article presents a novel sensor-fusion method solutions for vehicle tracking exploit the satellite navigation
for indoor vehicle tracking. The phase of the signals backscat- systems, for example, global positioning system (GPS) and
tered by a set of Ultra High Frequency-Radio Frequency Iden- other global navigation satellite systems (GNSSs).
tification (UHF-RFID) reference tags spread in the scenario is
combined with the information acquired by on-board low-cost However, the signal attenuation due to the building walls
kinematic sensors. The RFID data are acquired by the on-board hardly limits their employment in indoor scenarios. In this
reader, during the relative motion of the vehicle with respect context, several alternative solutions have been proposed to
to the static reference tags, by resembling a synthetic-array develop indoor positioning systems (IPSs) that are specific for
approach, with an advantageous reduction of the reference-tag autonomous vehicles [6].
spatial density. In particular, such phase samples are combined
with the kinematic data collected by odometers, through a A detailed state-of-the-art analysis discussing the main
sensor-fusion approach. The method capability is investigated features of the different technologies together with their critical
through a numerical analysis that accounts for the main sys- issues can be found in [7]. Classical solutions are based on
tem parameters. Then, the tracking capability is demonstrated computer vision, ultrasounds, magnetic fields, and laser imag-
through a measurement campaign in a laboratory test set with a ing detection and ranging (LIDAR) [7]–[11]. More recently, a
UHF-RFID robot prototype equipped with commercial encoders.
Experimental results show an average localization error of widespread variety of radio frequency (RF) technologies such
centimeter order in the estimation of medium-length trajectories as Wi-Fi, ZigBee, Bluetooth, UltraWide-Band (UWB), and RF
by employing only two reference tags in a relatively small area. identification (RFID) have been proposed [12]–[16]. Generally
The proposed method does not need for any calibration procedure speaking, radio waves are more robust to the presence of obsta-
and can be implemented by commercial off-the-shelf (COTS) cles and they can guarantee good performance in dynamic and
hardware.
crowded environments, such as shops or warehouses, where
Index Terms— Radio frequency identification (RFID) appli- the continuous motion of people or objects could interfere on
cations, RFID localization, RFID robot, RFID vehicle, robot the vehicle tracking based on LIDAR or cameras, the latter
localization, synthetic aperture radar, tag localization, ultra-
high frequency (UHF)-RFID position measurement, vehicle being also prone to privacy issues.
tracking. In such framework, passive RFID systems represent a
valuable solution, thanks to their low cost, easy deployment,
I. I NTRODUCTION and maintenance [17], [18]. Furthermore, if the vehicle is
already equipped with the RFID reader for inventory pur-
N OWADAYS, the localization of mobile devices, agents,
robots, and people has gained increasing attention to
develop solutions for the Internet of Things (IoT) and the
poses [19]–[21], it can also be exploited for the vehicle self-
localization, with no additional hardware and then reducing
Industrial IoT (IIoT) paradigms [1]–[3]. Indeed, location the total cost of the tracking infrastructure. When phase-based
awareness represents an enabling feature for a lot of social and methods are implemented [22], a RFID tracking system allows
industrial applications [4], [5]. In outdoor scenarios, typical to get a centimeter order localization error similar to the one
achieved by UWB nodes with reduced costs. Indeed, despite
Manuscript received June 22, 2020; accepted September 20, 2020. Date UWB devices allow to cover larger areas with less reference
of publication September 30, 2020; date of current version January 5, 2021. anchors, only additional passive RFID tags are required as
This work was supported in part by the University of Pisa within the Project
AURORA of the “Bando Dimostratori Tecnologici” (D.R 1528, September 13, passive beacons if the robot is already equipped with RFID
2018), in part by the Italian Ministry of Education and Research (MIUR) hardware for inventory purposes. They are low-cost devices,
in the framework of the CrossLab Project (Departments of Excellence), which do not require for batteries or any other powering
and in part by the University of Pisa under the PRA 2018–2019 Research
Project “CONCEPT—COmmunication and Networking for vehicular CybEr- sources. Thus, they can be easily installed all around the
Physical sysTems.” The Associate Editor coordinating the review process was application scenario, with a spatial density depending on the
Yuan Gao. (Corresponding author: Alice Buffi.) expected tracking performance. The analysis of the RFID-
Andrea Motroni and Paolo Nepa are with the Department of Infor-
mation Engineering, University of Pisa, Pisa 56122, Italy (e-mail: based tracking systems [23]–[43] is better described later in
andrea.motroni@ing.unipi.it; paolo.nepa@unipi.it). the next section.
Alice Buffi and Bernardo Tellini are with the Department of Energy, Recently, the authors of this article patented a novel
Systems, Territory and Constructions Engineering, University of Pisa, Pisa
56122, Italy (e-mail: alice.buffi@unipi.it; bernardo.tellini@unipi.it). RFID phase-based method for vehicle tracking [44]. The
Digital Object Identifier 10.1109/TIM.2020.3027926 method combines the phase of the signals backscattered by
1557-9662 © 2020 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See https://www.ieee.org/publications/rights/index.html for more information.
Authorized licensed use limited to: University of Pisa. Downloaded on August 29,2022 at 19:53:05 UTC from IEEE Xplore. Restrictions apply.
8001314 IEEE TRANSACTIONS ON INSTRUMENTATION AND MEASUREMENT, VOL. 70, 2021
the Ultra-High Frequency (UHF)-RFID reference tags with to the widespread utilization of such identification technology
some kinematic data, through sensor-fusion techniques. Phase that can guarantee much larger reading ranges (up to 10 m).
samples can be collected during the relative motion of the Beyond the tag detection information [25], [26], the ampli-
robot-mounted RFID antenna with respect to the reference tude [27] and the phase [22] of the tag backscattered signal
tags, by resembling a synthetic-array approach [45], [46]. can be both employed for localization purposes. The signal
A preliminary numerical analysis was proposed in [47]. amplitude is commonly measured through the received signal
The method is mainly proposed to face with the problem of strength indicator (RSSI). Performance of RSSI-based solu-
the robot self-localization, when the latter is already equipped tions depends on the typology of the reference tags, and they
with an RFID reader for inventory purposes. Thus, the cost require time-consuming calibration procedures, which should
of the tracking systems is mainly given by the reference-tag be repeated anytime some changes occur in the scenario.
deployment. In complex scenarios, where strong reflections On the other hand, phase-based localization methods typically
and obstructions may occur, due to the presence of metallic allow for a better localization accuracy, and they are more
shelves or other obstacles, the proposed UHF-RFID tracking robust to the multipath propagation phenomena with respect
system still guarantees good performance especially thanks to to the amplitude-based methods, as described in [28]. Phase-
the exploitation of the robot movement which significantly based localization requires a coherent RFID receiver, which
mitigates shadowing and multipath effects. On the contrary, is employed in most commercial off-the-shelf (COTS) UHF-
UWB systems operating at frequencies greater than 1 GHz RFID readers. It is noteworthy that phase measurements may
would require for complex multipath cancellation algorithms. be exploited only with tags operating through the modulated
Moreover, people continuously walking and moving goods backscattering principle, namely with passive or semipassive
might interfere with a laser-navigated robot. RFID tags [18].
The proposed method can be employed both in small sce- In phase-based UHF-RFID methods, two main issues exist.
narios, for example, retails, and in big scenarios, for example, The first one is the ambiguity due to the phase periodic-
huge warehouses, provided that the reference tags meet the ity, which is typically solved through the employment of
required performance in terms of sensitivity and detection operations with phasors [29], or unwrapping techniques [30],
range. [31]. The second one is the presence of a phase offset term,
This article presents the mathematical formulation of the which is unknown a priori and depends on the tag typology,
sensor-fusion method [44] when different kinematic sensors the reader antenna, the transmitted power level, and the
are used to estimate the 3-D trajectory of the vehicle. Besides, tag–reader relative orientation. To release from long calibration
a particular case is described, in which the RFID phase procedures, the phase difference of arrival (PDoA) approach
samples are combined with the kinematic data collected by can be employed.
odometers placed on a wheeled vehicle moving at the ground To reduce the number of reference tags when performing
side. The method is characterized through both a numerical vehicle tracking in relatively large spaces, RFID data can be
analysis as well as experimental results to show the tracking suitably combined with data from proprioceptive kinematic
capability in real scenarios. sensors, such as encoders, optical flow sensors, or Inertial
This article is organized as follows. Section II describes Measurement Units (IMUs) equipped with accelerometers,
a state-of-the-art analysis of the RFID-based systems for gyroscopes, and magnetometers. The data combination typ-
vehicle tracking. Section III provides a detailed description of ically occurs through a sensor-fusion approach [32], [33].
the tracking method by combining RFID data and kinematic Usually, data fusion algorithms rely on sequential Bayesian
data, together with the particular case of odometer adoption. or Monte Carlo estimators for dynamic systems, such as, for
In Section IV, the algorithm capability is investigated through example, the Kalman Filter and its extensions, for example,
numerical simulations by accounting for the most relevant Extended Kalman Filter (EKF) or Unscented Kalman Filter
algorithm parameters and geometrical quantities. In Section V, (UKF), and the Particle Filter (PF). In some cases, smoothing
an experimental validation is presented, through real mea- algorithms, or finite impulse response filters are adopted too.
surements in an indoor scenario by an RFID robot. Tracking Typically, two main algorithm steps can be recognized: a
performance is demonstrated through a comparison with a prediction step and an update step. The first one is usually
standard dead-reckoning algorithm. Finally, Section VI states executed through the kinematic sensor data, for example,
conclusions and future work. odometry data, and it consists of a first hypothesis on the
vehicle path. In detail, the moving agent trajectory is estimated
II. R ELATED W ORKS through successive integrations of kinematic data, namely with
The first tracking methods exploited the RFID system a dead-reckoning approach [34], [35]. After that, the RFID
working at the HF band (13.56 MHz) [23]. The mobile node is data are employed in the update step, to compensate for the
a robot equipped with the HF-RFID reader which detects the drift of the estimated trajectory due to the errors accumulated
reference tags deployed at the floor side. The robot position by the kinematic sensors.
can be estimated as the position of the detected reference tag, In [36] and [37], the mobile node, for example, a robot,
thus the localization resolution depends on the tag density at is equipped with an RFID reader, and fixed RFID reference
the floor. tags are deployed in the environment. Then, sensor-fusion
More recently, the employment of the RFID systems oper- techniques or estimation algorithms such as Kalman filter, PF,
ating in the UHF band (865–928 MHz) [24] is increasing, due or their variations are adopted for positioning estimation.
Authorized licensed use limited to: University of Pisa. Downloaded on August 29,2022 at 19:53:05 UTC from IEEE Xplore. Restrictions apply.
MOTRONI et al.: SENSOR-FUSION AND TRACKING METHOD FOR INDOOR VEHICLES 8001314
Authorized licensed use limited to: University of Pisa. Downloaded on August 29,2022 at 19:53:05 UTC from IEEE Xplore. Restrictions apply.
8001314 IEEE TRANSACTIONS ON INSTRUMENTATION AND MEASUREMENT, VOL. 70, 2021
Through the article, the subscript “k” indicates the k-th B. RFID-Based Tracking Method
value of the discrete sequence of the corresponding parameter Now, we consider a set of hypothetical positions P =
at the time tk . {1, . . . , N p } of the vehicle initial position p1 . We name those
Now, the vehicle position at the k-th instant pk can be as particles, by inspiring to the PF algorithms [44]. Thus, the
written as a displacement pk ∈ R3 relative to the position at (p)
position of the p-th particle at the initial instant, q1 ∈ R3 ,
the previous instant k − 1, according to can be modeled as
⎛ ( p) ⎞
pk = pk−1 + pk ∀k ∈ Wn . (1) ux
(p) ⎜ ( p) ⎟
q1 = p1 + H⎝ u y ⎠ ∀ p ∈ P (6)
It is noteworthy that the vehicle/moving agent position repre- ( p)
uz
sents the position of the RFID reader antenna phase center. All
( p) ( p) ( p)
around the scenario, NT passive tags are considered, which are where [u x , u y , u z ]T ∈ R3 is a vector of normal random
placed at known positions pTj = [x T j , yT j , z T j ]T ∈ R3 , being variables with zero mean value and unitary variance. Without
j ∈ WT , the tag index and WT = {1, . . . , NT } the set of tags. loss of generality, we assumed H ∈ R3×3 to be diagonal and
Along the path, the vehicle-mounted kinematic sensors defined as
measure the kinematic data with a time sampling of t, ⎛ ⎞
σx 0 0
by guaranteeing the local observability of the dynamic
H = ⎝ 0 σy 0 ⎠ (7)
system [43]. 0 0 σz
Similarly, the UHF-RFID reader collects the tag backscat-
tered signals, and without loss of generality, we suppose that where σx , σ y , and σz are the standard deviations of the initial
the tag replies to the reader queries at the same instants as position errors.
those of the kinematic data collection, namely with the same Starting from the set of N p particles, it is possible to build
time sampling of t. a set of particle trajectories, namely a set of hypothetical
Now, by considering the n-th portion of the vehicle trajec- trajectories of the vehicle. In particular, the p-th particle
(p)
tory, we suppose to collect a set Wn = {1, . . . , Nn } of both position at the k-th instant, qk ∈ R3 , can be derived as a
kinematic data and RFID phase data, being Nn the number function of the position at the previous step (k − 1) according
of collected samples. The selection of such parameter will be to:
⎛ ( p) ⎞
discussed later. Thus, along the n-th portion the index k ∈ Wn . ux
The phase of the signal backscattered by the j -th tag at the (p) (p) ⎜ ( p) ⎟
qk = qk−1 +pk + α ⎝ u y ⎠ ∀k ∈ Wn , ∀ p ∈ P (8)
k-th instant can be written as ( p)
uz
4πdk, j
ϕk, j = + ϕ0, j ∀k ∈ Wn , ∀ j ∈ WT (2) where α is empirically determined on the basis of the class of
λ
accuracy of the kinematic sensors and the speed of the vehicle,
where and pk ∈ R3 represents the kinematic-based displacement
of the vehicle position at the k-th instant with respect to the
dk, j = pk − pTj 2 ∀k ∈ Wn , ∀ j ∈ WT (3) previous one, affected by error. Such displacement can be
written as
is the distance between the vehicle position pk and the j -th ⎛ ⎞
tag position pTj , λ is the free-space wavelength, and ϕ0, j is εx k
the phase offset of the j -th tag, which includes the effect of pk = pk + ⎝ εyk ⎠ ∀k ∈ Wn (9)
cables, reader components, and the tag typology itself [27]. εzk
· 2 represents the Euclidean norm. where [εxk , εyk , εzk ]T ∈ R3 is the error vector on the
To release from the knowledge of the phase offset, we con- displacement of the vehicle position at the k-th instant, when
sider the phase difference between each sample and a reference considering measurements done by the kinematic sensors.
one, to construct the so-called relative phase history. Indeed, Thus, N p particle trajectories starting from the random
the above phase offset can be considered as constant if the (p) (p)
position q1 up to the position qNn are formed. They have
tag is detected within the main beam of the reader antenna different shapes due to the random component in (8), which
radiation pattern. Without loss of generality, we choose the is the process noise [49] necessary to emulate the effect of the
first available sample as reference (k = 1), thus the relative typical uncertainty growth over the time of the position suc-
phase history reads as follows: cessive updates by proprioceptive kinematic sensors. We could
imagine the particles as a cloud of points which moves
ϕ k, j = ϕ k, j − ϕ1, j ∀k ∈ Wn , ∀ j ∈ WT . (4)
and expands: more the cloud is wide, more the hypothetical
The term ϕ 1, j becomes equal to zero and we build the trajectories are widespread.
phasor sequence bT j ∈ C Nn for the j -th tag Now, from the particle trajectories it is possible to build a
(p)
set of hypothetical distances between the particle position qk
T and the j -th tag
bT j = 1, e− j ϕ 2, j , . . . , e− j ϕ k, j , . . . , e− j ϕ Nn , j
( p) (p)
∀ j ∈ WT . (5) d̃k, j = qk − pTj ∀k ∈ Wn , ∀ j ∈ WT , ∀ p ∈ P. (10)
2
Authorized licensed use limited to: University of Pisa. Downloaded on August 29,2022 at 19:53:05 UTC from IEEE Xplore. Restrictions apply.
MOTRONI et al.: SENSOR-FUSION AND TRACKING METHOD FOR INDOOR VEHICLES 8001314
From here, the hypothetical phase of the signal backscattered The number of consecutive displacements which allow to
by the j -th tag can be calculated (11) and consequently a get the traveled path higher than a fixed threshold L th , deter-
set of hypothetical relative phase histories (12) can be written mine the number Nn of the points of the set Wn . Now, the
according to (4) subset On ⊂ Wn is chosen in such a way to estimate the
( p) first part of length L est of the vehicle trajectory portion of
( p) 4π d̃k, j
φk, j = ∀k ∈ Wn , ∀ j ∈ WT , ∀ p ∈ P (11) length L th [Fig. 1(b)]. In other words, the tracking method
λ processes groups of data collected during the vehicle trajectory
( p) ( p) ( p)
φk, j = φk, j − φ1, j ∀k ∈ Wn , ∀ j ∈ WT , ∀ p ∈ P. (12) portion of length L th to estimate a trajectory portion of length
Thus, a set of N p hypothetical phasor sequences can be L est ≤ L th .
derived (13) The trajectory length threshold L th is a parameter that
T
affects the tracking performance. Indeed, the proposed RFID-
( p) ( p) ( p)
(p)
aT j = 1, e− j φ2, j , . . . , e− j φk, j , . . . , e− j φ Nn , j based method is a synthetic-array approach, which requires
a relative motion between the vehicle and the tags, and
∀k ∈ Wn , ∀ j ∈ WT , ∀ p ∈ P. (13)
exhibits a localization accuracy that improves for longer
Now, for the j -th detected tag, it is possible to compare the paths. On the other hand, longer spatial intervals L th may
phasor sequence (5), with the set of the hypothetical phasor improve the method accuracy at the expense of longer process-
sequences through a matching index (14) ing temporal intervals, especially for low-speed vehicles.
(p) This trade-off must be considered if real-time tracking is
bTj , aTj mandatory.
( p)
Cn, j = ∀ j ∈ WT , ∀ p ∈ P (14)
bT (p) Next, the estimation procedure is repeated by using a
j 2 aT j
2 new set of kinematic and RFID data, which are collected
being · the scalar product operator and the operator · 2 during a vehicle trajectory portion of length L th , consecutive
applied to a vector of complex numbers must be intended as with respect to the previous estimated portion of length
the Euclidean norm in the complex domain. Consequently, L est . If L est = L th , the processed data always belong to
( p)
Cn, j is a scalar real number in the [0, 1] interval. not-overlapped trajectory portions, otherwise data belong to
Then, for the p-th particle, we combine the matching overlapped trajectory portions.
( p)
indexes Cn, j obtained from all the detected tags ( j ∈ WT ) After determining the estimated vehicle position p̂k for
through an average operation according to (15): k ∈ On , we select a subset Sn ⊂ P of hypothetical trajectories
that show higher average matching indexes [50]. In particular,
1
NT
( p) ( p)
Cn = C ∀ p ∈ P. (15) we select a factor η ∈ (0, 1] of the particles with the best
NT j =1 n, j average matching index and we get a number of “survivor”
Since the matching index (14) is in the interval [0, 1], the aver- particles equal to Nsp in such a way that Np /Nsp = 1/η ∈
age matching index (15) belongs to the same interval. N+ . Before to proceed with a further portion estimation,
( p) to overcome the “sample impoverishment” phenomena [51],
The Cn is a kind of likelihood index which indicates if
it is required to replicate the selected Nsp particles of a factor
the p-th particle trajectory is close or not to the actual vehicle
( p) 1/η. In other words, each survivor particle is copied 1/η times.
trajectory. In fact, a set of similar d̃k, j values with respect to
( p) Therefore, we can start the consecutive estimation procedure
the dk, j values leads to a set of φk, j close to the ϕ k, j , thus with a set of N p particles, by keeping unchanged the particle
( p)
resulting in a high value of the average matching index Cn . number employed in the n-th trajectory portion estimation. The
( p)
After determining the Cn indexes for all the particles, block diagram of the proposed tracking method is reported
they are employed to estimate the 3-D vehicle position in Fig. 2.
p̂k = [x̂ k , ŷk , ẑ k ] at the k-th step through a weighted average It is noteworthy that, due to the anticollision protocol of
operation: the EPC Global C1 G2 [52], the tag replies occur at different
( p) time slots. To account for such phenomena, the algorithm
(p)
p∈P Cn qk should be applied by considering a subset Wn j of the set Wn
p̂k = ( p) ∀k ∈ On (16)
which is different for each j -th tag. The reference reading
p∈P C n
to build the relative phase history becomes the first available
where On ⊂ Wn is a subset of the set of the processed number reading of each subset Wn j . This peculiarity makes possible to
of RFID and kinematic readings. run the algorithm even in large areas with a low-density tag
In more detail, first, we need to determine a selection deployment, once a few successful readings or intermitting
criterion of the set Wn . During the vehicle movement, it is readings are available. We recommend the tag deployment in
possible to estimate the pk vehicle displacement from the any position to ensure that at least one RFID tag is inside
kinematic sensor measurements, as better described in the next the robot antenna detection area for each trajectory portion.
section. By summing such consecutive displacements, it is Besides, we recommend to carefully choose the tag chip
possible to derive the vehicle traveled path at the k-th instant which achieves the detection range required by the application
scenarios, being either a small retail shop or a huge warehouse.
k
Lk = ph . (17) If it is not possible, a slight performance degradation must be
2
h=1 accepted.
Authorized licensed use limited to: University of Pisa. Downloaded on August 29,2022 at 19:53:05 UTC from IEEE Xplore. Restrictions apply.
8001314 IEEE TRANSACTIONS ON INSTRUMENTATION AND MEASUREMENT, VOL. 70, 2021
Authorized licensed use limited to: University of Pisa. Downloaded on August 29,2022 at 19:53:05 UTC from IEEE Xplore. Restrictions apply.
MOTRONI et al.: SENSOR-FUSION AND TRACKING METHOD FOR INDOOR VEHICLES 8001314
Authorized licensed use limited to: University of Pisa. Downloaded on August 29,2022 at 19:53:05 UTC from IEEE Xplore. Restrictions apply.
8001314 IEEE TRANSACTIONS ON INSTRUMENTATION AND MEASUREMENT, VOL. 70, 2021
Fig. 5. Simulated “S-shaped” actual and estimated trajectories with a reference-tag grid step of s = 5 m, by considering data acquired along the
(a) n = 1 portion, the (b) n = 15 portion, the (c) n = 27 portion, the (d) n = 55 portion, the (e) n = 80 portion, and the (f) n = 116 portion.
Authorized licensed use limited to: University of Pisa. Downloaded on August 29,2022 at 19:53:05 UTC from IEEE Xplore. Restrictions apply.
MOTRONI et al.: SENSOR-FUSION AND TRACKING METHOD FOR INDOOR VEHICLES 8001314
Fig. 8. Simulated localization errors for two tracking methods by considering the “S-shaped” vehicle trajectory of Fig. 5 with a reference-tag grid step of
s = 5 m. (a) Error on the x-coordinate. (b) Error on the y-coordinate. (c) Residual cartesian error.
Fig. 9. Simulated localization errors for the proposed sensor-fusion tracking method by considering the “S-shaped” vehicle trajectory of Fig. 5 by varying
the reference-tag grid step of s. (a) Error on the x-coordinate. (b) Error on the y-coordinate. (c) Residual cartesian error.
Fig. 10. Histogram of localization errors for two tracking methods by the “S-shaped” trajectory of Fig. 5 for ten different Monte Carlo simulations with
s = 5 m and L th = 1 m. (a) Error on the x-coordinate. (b) Error on the y-coordinate. (c) Residual cartesian error.
sensor-fusion tracking method with s = 5 m and L th = When increasing the length L th , the localization error
1 m (green bars) to the “S-shaped” trajectory of Fig. 3. reduces, as shown by the relative frequency bars. Indeed, when
The results are related to 6000 positions along the trajectory the RFID data are acquired and processed along a longer path,
and ten repeated Monte Carlo simulations. A path step of the average matching indexes (15) of the considered particle
10 cm is employed in the graphs; thus, a main bar can be trajectories are more distant each other. This means that such
noticed for the sensor-fusion tracking method, which reaches parameter is more significant to estimate the vehicle trajectory
nearby the 100% relative frequency. On the contrary, the error for longer trajectory portions, according to the synthetic-array
distributions of the dead-reckoning approach are in the interval theory [29].
[−5, 5] m for the x-coordinate and the y-coordinate, while the
residual cartesian error is in the interval [0, 7] m. C. Effect of the Sensor Accuracy
The class of accuracy of the kinematic sensors impacts on
B. Effect of the Portion Length Threshold L th the method performance.
The effect of the length of the trajectory portion threshold We expect that larger errors of the rotary encoders lead to
L th on the tracking performance was also investigated. We still a lower accuracy for the sensor-fusion tracking method.
consider the “S-shaped” trajectory of Fig. 3 with the reference- To verify such effect, we investigated the impact of the
tag grid step of s = 5 m. Fig. 11 shows the histogram encoder resolution on the dead reckoning and the sensor-fusion
of the localization error on the x-coordinate [Fig. 11(a)], the tracking method, when the robot moves along a “C-shape”
y-coordinate [Fig. 11(b)], and the histogram of the residual trajectory of length 4 m (Fig. 12) similar to that acquired
cartesian error [Fig. 11(c)], when the trajectory length thresh- during the experimental analysis (Section V).
old is equal to L th = 1 m (blue bars), L th = 2 m (orange bars), The robot had an average speed of 0.16 m/s, and
and L th = 3 m (green bars). The histogram still considers N R = 1696 data samples were acquired by the encoders
the localization results of 6000 positions of the “S-shaped” along the whole trajectory. Two reference RFID tags were
trajectory of Fig. 3, for ten repeated Monte Carlo simulations. attached at the ceiling at pT1 = [1.00, 0.65, 2.40]T m and
Authorized licensed use limited to: University of Pisa. Downloaded on August 29,2022 at 19:53:05 UTC from IEEE Xplore. Restrictions apply.
8001314 IEEE TRANSACTIONS ON INSTRUMENTATION AND MEASUREMENT, VOL. 70, 2021
Fig. 11. Histogram of localization errors versus L th variations for the proposed method by considering the “S-shaped” trajectory of Fig. 5 for ten different
Monte Carlo simulations with s = 5 m. (a) Error on the x-coordinate. (b) Error on the y-coordinate. (c) Residual cartesian error.
Authorized licensed use limited to: University of Pisa. Downloaded on August 29,2022 at 19:53:05 UTC from IEEE Xplore. Restrictions apply.
MOTRONI et al.: SENSOR-FUSION AND TRACKING METHOD FOR INDOOR VEHICLES 8001314
Fig. 13. Average simulated localization errors for the proposed sensor-fusion tracking method versus encoder quantization interval qe by considering
N R = 1696 positions of the “C-shape” trajectory of Fig. 12 for ten different Monte Carlo simulations. (a) Error on the x-coordinate. (b) Error on the
y-coordinate. (c) Residual cartesian error.
Fig. 17. Measured “C-shape” trajectory by the laser range finder and
estimated trajectories by measured data for two tracking methods, when two
reference tags are considered in the scenario.
Fig. 15. Experimental indoor scenario with the UHF-RFID robot and the
two reference tags. with the LRF data (solid black line). It can be noticed that the
RFID-based estimated trajectory is closer to the LRF ground
truth than the trajectory estimated by the dead-reckoning
The fluctuation on the curve behavior suggests the presence approach. It is worth noting that the RFID-based estimated
of strong multipath components since the antenna has a trajectory obtained by processing the measured data is different
considerably wide HPBW. To perform the robot tracking, with respect to that obtained by processing the simulated
the proposed sensor-fusion method was applied, with the fol- data and illustrated in Fig. 12. Such differences, even if
lowing parameters: L th = 2.5 m, L est = 0.3 m, [σx , σ y , σz ] = minimal, can be seen at the first trajectory corner and at
[0.1, 0.1, 0] m, N p = 500, α = 5 × 10−4 , η = 0.25. the trajectory end. The similarity of the two abovementioned
The particles were randomly initialized around the assumed estimated trajectories confirms the validity of the simulation
vehicle initial position p1 . Fig. 17 shows the estimated tra- model including the effect of thermal noise and multipath on
jectory (dashed green line) by applying the proposed sensor- the phase samples.
fusion tracking method to the measured data in Fig. 16. The To better visualize the result of Fig. 17, Fig. 18 shows
trajectory estimated by applying a dead-reckoning approach the localization errors on the vehicle position estimation with
(dash-dotted red line) is also shown together with the trajectory respect to the LRF ground truth of Fig. 17. The proposed
estimated by employing a state-of-the-art SLAM algorithm RFID-based tracking method achieves a localization error
Authorized licensed use limited to: University of Pisa. Downloaded on August 29,2022 at 19:53:05 UTC from IEEE Xplore. Restrictions apply.
8001314 IEEE TRANSACTIONS ON INSTRUMENTATION AND MEASUREMENT, VOL. 70, 2021
Fig. 18. Measured localization error with respect to the ground truth related to the estimated vehicle trajectories of Fig. 17. (a) Error on the x-coordinate.
(b) Error on the y-coordinate. (c) Residual cartesian error.
Fig. 19. Histogram of localization errors from all the nine measurement sessions. (a) x-coordinate. (b) y-coordinate. (c) Residual cartesian error.
Authorized licensed use limited to: University of Pisa. Downloaded on August 29,2022 at 19:53:05 UTC from IEEE Xplore. Restrictions apply.
MOTRONI et al.: SENSOR-FUSION AND TRACKING METHOD FOR INDOOR VEHICLES 8001314
[2] J. M. Batalla, C. X. Mavromoustakis, G. Mastorakis, N. N. Xiong, and [22] P. V. Nikitin, R. Martinez, S. Ramamurthy, H. Leland, G. Spiess, and
J. Wozniak, “Adaptive positioning systems based on multiple wireless K. V. S. Rao, “Phase based spatial identification of UHF RFID tags,”
interfaces for industrial IoT in harsh manufacturing environments,” in Proc. IEEE Int. Conf. RFID (IEEE RFID), Orlando, FL, USA,
IEEE J. Sel. Areas Commun., vol. 38, no. 5, pp. 899–914, May 2020, Apr. 2010, pp. 102–109, doi: 10.1109/RFID.2010.5467253.
doi: 10.1109/JSAC.2020.2980800. [23] P. Ki-Cheol, P. San-Heon, and K. Won-Jun, “Apparatus and method
[3] F. Shamsfakhr, L. Palopoli, D. Fontanelli, A. Motroni, and of recognizing position and direction of mobile robot,” U.S.
A. Buffi, “Robot localisation using UHF-RFID tags for Patent 2 003 236 590 A1, Dec. 25, 2003.
industrial IoT applications,” in Proc. IEEE Int. Workshop [24] T. Deyle, H. Nguyen, M. S. Reynolds, and C. C. Kemp, “RFID-guided
Metrol. Ind. 4.0 IoT, Roma, Italy, Jun. 2020, pp. 659–664, robots for pervasive automation,” IEEE Pervas. Comput., vol. 9, no. 2,
doi: 10.1109/MetroInd4.0IoT48571.2020.9138210. pp. 37–45, Apr. 2010, doi: 10.1109/MPRV.2010.17.
[4] A. Yassin et al., “Recent advances in indoor localization: A sur- [25] L. Yang, J. Cao, W. Zhu, and S. Tang, “Accurate and efficient object
vey on theoretical approaches and applications,” IEEE Commun. tracking based on passive RFID,” IEEE Trans. Mobile Comput., vol. 14,
Surveys Tuts., vol. 19, no. 2, pp. 1327–1346, 2nd Quart., 2017, no. 11, pp. 2188–2200, Nov. 2015, doi: 10.1109/TMC.2014.2381232.
doi: 10.1109/COMST.2016.2632427. [26] A. Koch and A. Zell, “RFID-enabled location fingerprinting based
[5] K. Finkenzeller, RFID Handbook: Radio-Frequency Identification Fun- on similarity models from probabilistic similarity measures,” in Proc.
damentals and Applications. Hoboken, NJ, USA, Wiley, 2004. IEEE Int. Conf. Robot. Autom. (ICRA), Stockholm, Sweden, May 2016,
[6] V. Renaudin et al., “Evaluating indoor positioning systems in pp. 4557–4563, doi: 10.1109/ICRA.2016.7487656.
a shopping mall: The lessons learned from the IPIN 2018 [27] S. S. Saab and Z. S. Nakad, “A standalone RFID indoor positioning
competition,” IEEE Access, vol. 7, pp. 148594–148628, 2019, system using passive tags,” IEEE Trans. Ind. Electron., vol. 58, no. 5,
doi: 10.1109/ACCESS.2019.2944389. pp. 1961–1970, May 2011, doi: 10.1109/TIE.2010.2055774.
[7] L. Angrisani, P. Arpaia, and D. Gatti, “Analysis of localization [28] M. B. Akbar, D. G. Taylor, and G. D. Durgin, “Amplitude and phase
technologies for indoor environment,” in Proc. IEEE Int. Work- difference estimation bounds for multisensor based tracking of RFID
shop Meas. Netw. (M&N), Naples, Italy, Sep. 2017, pp. 1–5, tags,” in Proc. IEEE Int. Conf. RFID (RFID), San Diego, CA, USA,
doi: 10.1109/IWMN.2017.8078385. Apr. 2015, pp. 105–112, doi: 10.1109/RFID.2015.7113080.
[8] D. Fontanelli, F. Moro, T. Rizano, and L. Palopoli, “Vision-based robust [29] A. Buffi, A. Motroni, P. Nepa, B. Tellini, and R. Cioni, “A SAR-based
path reconstruction for robot control,” IEEE Trans. Instrum. Meas., measurement method for passive-tag positioning with a flying UHF-
vol. 63, no. 4, pp. 826–837, Apr. 2014, doi: 10.1109/TIM.2013.2289091. RFID reader,” IEEE Trans. Instrum. Meas., vol. 68, no. 3, pp. 845–853,
[9] H.-H. Lin, C.-C. Tsai, and J.-C. Hsu, “Ultrasonic localization and pose Mar. 2019, doi: 10.1109/TIM.2018.2857045.
tracking of an autonomous mobile robot via fuzzy adaptive extended [30] A. Tzitzis et al., “Localization of RFID tags by a moving
information filtering,” IEEE Trans. Instrum. Meas., vol. 57, no. 9, robot, via phase unwrapping and non-linear optimization,” IEEE J.
pp. 2024–2034, Sep. 2008, doi: 10.1109/TIM.2008.919020. Radio Freq. Identificat., vol. 3, no. 4, pp. 216–226, Dec. 2019,
[10] F. Santoni, A. De Angelis, I. Skog, A. Moschitta, and P. Carbone, doi: 10.1109/JRFID.2019.2936969.
“Calibration and characterization of a magnetic positioning system [31] C. Li et al., “RePos: Relative position estimation of UHF-RFID tags for
using a robotic arm,” IEEE Trans. Instrum. Meas., vol. 68, no. 5, item-level localization,” in Proc. IEEE Int. Conf. RFID Technol. Appl.
pp. 1494–1502, May 2019, doi: 10.1109/TIM.2018.2885590. (RFID-TA), Pisa, Italy, Sep. 2019, pp. 357–361, doi: 10.1109/RFID-
[11] P. Tripicchio, M. Satler, M. Unetti, and C. A. Avizzano, “Con- TA.2019.8892174.
fined spaces industrial inspection with micro aerial vehicles and laser [32] D. Smith and S. Singh, “Approaches to multisensor data fusion in target
range finder localization,” Int. J. Micro Air Vehicles, vol. 10, no. 2, tracking: A survey,” IEEE Trans. Knowl. Data Eng., vol. 18, no. 12,
pp. 207–224, Jun. 2018, doi: 10.1177/1756829318757471. pp. 1696–1710, Dec. 2006, doi: 10.1109/TKDE.2006.183.
[12] Y. Gao, H. Chen, Y. Li, C. Lyu, and Y. Liu, “Autonomous [33] D. Macii, A. Colombo, P. Pivato, and D. Fontanelli, “A data
Wi-Fi relay placement with mobile robots,” IEEE/ASME Trans. fusion technique for wireless ranging performance improvement,”
Mechatronics, vol. 22, no. 6, pp. 2532–2542, Dec. 2017, IEEE Trans. Instrum. Meas., vol. 62, no. 1, pp. 27–37, Jan. 2013,
doi: 10.1109/TMECH.2017.2751149. doi: 10.1109/TIM.2012.2209918.
[13] V. Bianchi, P. Ciampolini, and I. De Munari, “RSSI-based indoor [34] Y. Zhou and G. S. Chirikjian, “Probabilistic models of dead-
localization and identification for ZigBee wireless sensor networks in reckoning error in nonholonomic mobile robots,” in Proc. IEEE
smart homes,” IEEE Trans. Instrum. Meas., vol. 68, no. 2, pp. 566–575, Int. Conf. Robot. Autom., Taipei, Taiwan, Sep. 2003, pp. 1594–1599,
Feb. 2019, doi: 10.1109/TIM.2018.2851675. doi: 10.1109/ROBOT.2003.1241822.
[14] G. Michau, A. Nantes, A. Bhaskar, E. Chung, P. Abry, and P. Borgnat, [35] J.-X. Yu, Z.-X. Cai, and Z.-H. Duan, “Dead reckoning of mobile robot
“Bluetooth data in an urban context: Retrieving vehicle trajectories,” in complex terrain based on proprioceptive sensors,” in Proc. Int. Conf.
IEEE Trans. Intell. Transp. Syst., vol. 18, no. 9, pp. 2377–2386, Mach. Learn. Cybern., Kunming, China, Jul. 2008, pp. 1930–1935,
Sep. 2017, doi: 10.1109/TITS.2016.2642304. doi: 10.1109/ICMLC.2008.4620722.
[15] F. Lazzari, A. Buffi, P. Nepa, and S. Lazzari, “Numerical investigation of [36] R. Liu, G. Huskic, and A. Zell, “Dynamic objects tracking with a
an UWB localization technique for unmanned aerial vehicles in outdoor mobile robot using passive UHF RFID tags,” in Proc. IEEE/RSJ Int.
scenarios,” IEEE Sensors J., vol. 17, no. 9, pp. 2896–2903, May 2017, Conf. Intell. Robots Syst., Chicago, IL, USA, Sep. 2014, pp. 4247–4252,
doi: 10.1109/JSEN.2017.2684817. doi: 10.1109/IROS.2014.6943161.
[16] J. Zhang et al., “Robust RFID based 6-DoF localization for [37] J. J. Pomarico-Franquiz and Y. S. Shmaliy, “Accurate self-
unmanned aerial vehicles,” IEEE Access, vol. 7, pp. 77348–77361, 2019, localization in RFID tag information grids using FIR filtering,” IEEE
doi: 10.1109/ACCESS.2019.2922211. Trans. Ind. Informat., vol. 10, no. 2, pp. 1317–1326, May 2014,
[17] C.-H. Huang, L.-H. Lee, C. C. Ho, L.-L. Wu, and Z.-H. Lai, doi: 10.1109/TII.2014.2310952.
“Real-time RFID indoor positioning system based on Kalman- [38] S. Sarkka, V. V. Viikari, M. Huusko, and K. Jaakkola, “Phase-based
filter drift removal and heron-bilateration location estimation,” IEEE UHF RFID tracking with nonlinear Kalman filtering and smooth-
Trans. Instrum. Meas., vol. 64, no. 3, pp. 728–739, Mar. 2015, ing,” IEEE Sensors J., vol. 12, no. 5, pp. 904–910, May 2012,
doi: 10.1109/TIM.2014.2347691. doi: 10.1109/JSEN.2011.2164062.
[18] C. Li, L. Mo, and D. Zhang, “Review on UHF RFID localization [39] E. DiGiampaolo and F. Martinelli, “Mobile robot localization using the
methods,” IEEE J. Radio Freq. Identificat., vol. 3, no. 4, pp. 205–215, phase of passive UHF RFID signals,” IEEE Trans. Ind. Electron., vol. 61,
Dec. 2019, doi: 10.1109/JRFID.2019.2924346. no. 1, pp. 365–376, Jan. 2014, doi: 10.1109/TIE.2013.2248333.
[19] Fornes. (Feb. 28, 2017). Robots Will Be In Retail Stores Sooner Than [40] F. Martinelli, “A robot localization system combining RSSI and phase
You Think. [Online]. Available: https://www.forbes.com/sites/nikkibaird/ shift in UHF-RFID signals,” IEEE Trans. Control Syst. Technol., vol. 23,
2017/02/28/robots-in-retail-stores-closer-than-you-think/#1edac3ea5d96 no. 5, pp. 1782–1796, Sep. 2015, doi: 10.1109/TCST.2014.2386777.
[20] F. Bernardini et al., “Particle swarm optimization in SAR-based [41] F. Martinelli, “Simultaneous localization and mapping using the phase
method enabling real-time 3D positioning of UHF-RFID tags,” of passive UHF-RFID signals,” J. Intell. Robotic Syst., vol. 94, nos. 3–4,
IEEE J. Radio Freq. Identificat., early access, Jun. 30, 2020, pp. 711–725, Jun. 2019, doi: 10.1007/s10846-018-0903-8.
doi: 10.1109/JRFID.2020.3005351. [42] V. Magnago et al., “Robot localisation based on phase measures of
[21] F. Bernardini et al., “Robot-based indoor positioning of UHF-RFID backscattered UHF-RFID signals,” in Proc. IEEE Int. Instrum. Meas.
tags: The SAR method with multiple trajectories,” IEEE Trans. Instrum. Technol. Conf. (I2MTC), Auckland, New Zealand, May 2019, pp. 1–6,
Meas., doi: 10.1109/TIM.2020.3033728. doi: 10.1109/I2MTC.2019.8826864.
Authorized licensed use limited to: University of Pisa. Downloaded on August 29,2022 at 19:53:05 UTC from IEEE Xplore. Restrictions apply.
8001314 IEEE TRANSACTIONS ON INSTRUMENTATION AND MEASUREMENT, VOL. 70, 2021
[43] V. Magnago et al., “Ranging-free UHF-RFID robot position- Dr. Buffi is a member of the IEEE Instrumentation and Measurement Society
ing through phase measurements of passive tags,” IEEE Trans. and of the Antennas and Propagation Society. She was a recipient of the Best
Instrum. Meas., vol. 69, no. 5, pp. 2408–2418, May 2020, Paper Award at the IEEE RFID-TA 2019 International Conference and of
doi: 10.1109/TIM.2019.2960900. the Young Scientist Award from the International Union of Radio Science,
[44] A. Buffi, P. Nepa, A. Motroni, and B. Tellini, “Mobile device self- Commission B, in 2013 and 2016. She serves as an Associate Editor for the
location method using at least one passive radio-frequency device,” Appl. IEEE J OURNAL OF R ADIO F REQUENCY I DENTIFICATION.
no. EP2019 7 281.7, 2019.
[45] A. Motroni et al., “SAR-based indoor localization of UHF-RFID
tags via mobile robot,” in Proc. Int. Conf. Indoor Position-
ing Indoor Navigat. (IPIN), Nantes, France, Sep. 2018, pp. 1–8,
doi: 10.1109/IPIN.2018.8533847.
[46] A. Parr, R. Miesen, F. Kirsch, and M. Vossiek, “A novel method for
UHF RFID tag tracking based on acceleration data,” in Proc. IEEE
Int. Conf. RFID (RFID), Orlando, FL, USA, Apr. 2012, pp. 110–115,
doi: 10.1109/RFID.2012.6193037.
[47] A. Motroni, P. Nepa, A. Buffi, and B. Tellini, “A phase-based method
for mobile node localization through UHF-RFID passive tags,” in Proc.
Paolo Nepa (Senior Member, IEEE) received the
IEEE Int. Conf. RFID Technol. Appl. (RFID-TA), Pisa, Italy, Sep. 2019,
Laurea (Doctor) degree (summa cum laude) in elec-
pp. 470–475, doi: 10.1109/RFID-TA.2019.8892264.
tronics engineering from the University of Pisa, Pisa,
[48] S. Jain, S. Nandy, R. Ray, and S. N. Shome, “Application of particle fil-
Italy, in 1990.
tering technique for sensor fusion in mobile robotics,” in Proc. IEEE Int.
Since 1990, he has been with the Department of
Conf. Mechatronics Autom., Beijing, China, Aug. 2011, pp. 2285–2290,
Information Engineering, University of Pisa, where
doi: 10.1109/ICMA.2011.5986341.
he is currently a Full Professor. In 1998, he was with
[49] P. Vorst and A. Zell, “Particle filter-based trajectory estimation with
the Electro Science Laboratory (ESL), The Ohio
passive UHF RFID fingerprints in unknown environments,” in Proc.
State University (OSU), Columbus, OH, USA, as a
IEEE/RSJ Int. Conf. Intell. Robots Syst., St. Louis, MO, USA, Oct. 2009,
Visiting Scholar supported by a Grant of the Italian
pp. 395–401, doi: 10.1109/IROS.2009.5354627.
National Research Council. At the ESL, he was
[50] M. S. Arulampalam, S. Maskell, N. Gordon, and T. Clapp, “A tutorial
involved in research on efficient hybrid techniques for the analysis of large
on particle filters for online nonlinear/non-Gaussian Bayesian tracking,”
antenna arrays. In the context of UHF-RFID systems, he is working on
IEEE Trans. Signal Process., vol. 50, no. 2, pp. 174–188, Feb. 2002,
techniques for radiolocalization of either tagged objects or readers. He has
doi: 10.1109/78.978374.
coauthored more than 300 international journal articles and conference
[51] P. Vadakkepat and L. Jing, “Improved particle filter in sensor fusion for
contributions. His research interests include the extension of high-frequency
tracking randomly moving object,” IEEE Trans. Instrum. Meas., vol. 55,
techniques to electromagnetic scattering from material structures and its
no. 5, pp. 1823–1832, Oct. 2006, doi: 10.1109/TIM.2006.881569.
application to the development of radio propagation models for indoor and
[52] (2018). EPC UHF Gen2 Air Interface Protocol Standard. [Online].
outdoor scenarios of wireless communication systems. He is also involved
Available: https://www.gs1.org/epcrfid/epc-rfid-uhf-air-interface-
in the design of wideband and multiband antennas, mainly for base stations
protocol/2-0-1
and mobile terminals of communication systems, as well as in the design
of antennas optimized for near-field coupling and focusing. He was working
on channel characterization, wearable antenna design, and diversity scheme
Andrea Motroni (Member, IEEE) received the B.E. implementation, for body-centric communication systems.
and M.E. degrees (summa cum laude) in telecommu- Dr. Nepa is a member of the Technical Advisory Board of URSI Com-
nications engineering from the University of Pisa, mission B—Fields and Waves and served as the General Chair for the IEEE
Pisa, Italy, in 2015 and 2017, respectively, where RFID-TA 2019 International Conference. He was a recipient of the Young
he is pursuing the Ph.D. degree in information Scientist Award from the International Union of Radio Science, Commission
engineering. B, in 1998. Since 2016, he has been serving as an Associate Editor for the
He has coauthored one European patent applica- IEEE A NTENNAS AND W IRELESS P ROPAGATION L ETTERS .
tion. His current research interests include indoor
radiolocalization systems, with specific focus on
UHF-RFID phase-based algorithms for robot track-
ing, UHF-RFID smart gates and applications for
Internet of Things and Industry 4.0. In 2019, he has been the President of the
IEEE Student Branch of the University of Pisa.
Alice Buffi (Member, IEEE) received the B.S. Bernardo Tellini (Senior Member, IEEE) was born
and M.S. degrees (summa cum laude) in telecom- in Pisa, Italy, in 1969. He received the M.S. (Laurea)
munications engineering and the Ph.D. degree and Ph.D. degrees in electrical engineering from
(Doctor Europaeus) in “Applied electromagnetism the University of Pisa, Pisa, in 1993 and 1999,
in electrical and biomedical engineering, electronics, respectively.
smart sensors, nanotechnologies” from the Univer- Since 2000, he has been with the University
sity of Pisa, Pisa, Italy, in 2006, 2008, and 2012, of Pisa, where he is currently a Full Professor of
respectively. Electrical Measurements with the Department of
In 2011, she was a visiting Ph.D. student with the Energy, Systems, Territory and Construction Engi-
Queen Mary University of London, London, U.K. neering. His current research interests include char-
Since 2012, she has been with the University of acterization of electrical and magnetic properties of
Pisa, where she is currently an Associate Professor with the Department of materials, aging process in battery cells, electromagnetic emission associated
Energy, Systems, Territory and Construction Engineering. She has coauthored with fast electromagnetic transients, sensors and transducers for pulsed power
several international journal papers and international conferences contributions applications measurement of position, magneto hydrodynamic problems.
and one European patent and one European patent application. Her current Dr. Tellini is a member of the IEEE Instrumentation and Measurement
research topics include measurement methods to locate static or moving Society and of the IEEE Magnetics Society. He is currently the Chair of the
items through radio frequency identification (RFID) systems operating at the IEEE Italy Section. He was the Chair of European Pulsed Power Laboratories
ultra-high-frequency (UHF) band in Industry 4.0 scenarios. Besides, she has from 2010 to 2014. He served as the General Chair and the General Co-Chair
interests in classification methods for smart gates and smart storage systems for several IEEE Conferences in the field of measurements and other IEEE
and aging process in battery cells. technological topics.
Authorized licensed use limited to: University of Pisa. Downloaded on August 29,2022 at 19:53:05 UTC from IEEE Xplore. Restrictions apply.