A ROS Gazebo Plugin Design to Simulate RFID Systems

Simulation, robotics, and Radio Frequency Identification (RFID) technology have significant roles in the new industrial revolution and their applications are key aspects of making Industry 4.0 a reality. Developing efficient use cases in Industry 4.0 almost always requires accurate simulation tools to be used in the digital world. The problem of simulating RFID readers for robotics in environments where high populations of RFID tags exist is addressed in this paper. This paper will discuss the design of an RFID system plugin based on Robot Operating System (ROS) and Gazebo simulator and the probability-based model on which the plugin is based. To assess the performance of the proposed system model, the simulation results of the designed plugin are compared with experiments. We also prove that the proposed simulator is flexible enough to be used on any robot platform, including aerial and ground robots. We show initial results of the simulation of having an Unmanned Aerial Vehicle (UAV) and a Unmanned Ground Vehicle (UGV) equipped with an RFID reader, navigating in an environment in which RFID tags have been placed. The robots will be reading tags in different map layouts using RFID antennas, with different orientations. We compare the simulation and experimental results in terms of the total unique tag readings vs. time, for various map-layouts. Finally, we show how this plugin can be used in robotics research by using it to simulate a novel, RFID-based stigmergic navigation strategy. We illustrate, the accurate navigation of the UAV using the proposed plugin.


19
Robotics and automation are quickly becoming one of the 20 main success factor in e-commerce and Industry 4.0. They 21 have a very big impact on the world of logistics. The 22 number of multipurpose industrial robots developed and 23 designed in the Industry 4.0 for Europe alone has almost 24 doubled since 2004 [1]. An essential aspect of Industry The associate editor coordinating the review of this manuscript and approving it for publication was Nikhil Padhi . need to isolate its working area, its integration into human 30 workspaces becomes more economical and productive, and 31 opens up many possible applications in the industry [2]. 32 Industry 4.0 technology intends among other goals to revolu-33 tionize Inventory Management. New technologies are already 34 transforming how businesses is approaching inventory man-35 agement. AI algorithms, IoT-powered tracking systems and 36 robots can optimize existing inventory management pro-37 cesses and streamline business planning [3]. Simulations 38 has an important role in Industry 4.0. It leverages real-39 time data to mirror the physical world in a virtual model, 40 which can include robots, products, humans, and entire ware-41 houses. This allows operators to test and optimize the envi-42 ronment and robot configuration in the virtual world before 43 RF/analog design of the RFID reader. The model presented 99 by Han et al. models the signal generation in the reader 100 to verify whether the signal transmitted complies with the 101 specified spectrum mask in the radio regulations. There is 102 also a detailed model of the receiver part of the RFID reader 103 that further analyses the effect of transmitter/receiver cou-104 pling. The wireless channel in their simulator is modeled as 105 the vector addition of various multipaths. Authors in [16], 106 present an RFID simulation engine, called RFIDSim, which 107 implements the ISO 18000-6C communication protocol [17] 108 and supports path loss, fading, backscatter, capture, and tag 109 mobility models. They show that RFIDsim can be used to 110 simulate large populations featuring thousands of RFID tags. 111 Their model also simulates the deep fades that lead to fre-112 quent power losses of the passive RFID tags by modeling 113 the multipath effects statistically. RFIDSim aimed to facili-114 tate the relative comparison of different transmission control 115 strategies. An approach in [18], similar to RFIDsim, proposed 116 a simulation platform that relies on a discrete event simula-117 tor, designed also to implement a part of the ISO 18000-6C 118 communication protocol supporting path loss, backscatter, 119 capture, and tag mobility models. These models however are 120 either too old, hard to adapt to robotics simulation platforms, 121 or no longer available, and more importantly focused on 122 the low-level communication issues between tags and reader 123 antennas. Only a few of these models allow environment 124 remodeling, design or manipulation in real time. Most are 125 not open source. Finally, none of the simulators can be easily 126 adapted to work with robots or the tools that comes with 127 ROS. During the development of the proposed simulator in 128 this paper, a study was published by the authors in [19]. The 129 authors propose a simulator that is implemented as a Gazebo 130 plugin integrated with ROS. A tag localization algorithm 131 that uses the phase unwrapping technique and hyperbolae 132 intersection method employing a reader antenna mounted on 133 a mobile robot is used to estimate the position of the tags 134 deployed in the presented scenarios. The user needs to specify 135 the frequency, the range, the phase noise, and the gain of the 136 tag antenna, as well as physical parameters, like damping 137 coefficient and friction. The outcomes of their experiments 138 showed realistic results for environments with a low number 139 of tags (up-to 10). However, it is not known how the model 140 will behave when simulating real environments with large 141 populations featuring thousands of RFID tags. The illustrated 142 experiments comparing the simulator's performance with real 143 life do not test the behavior of the simulator using different 144 types of robots, asses the accuracy of RFID tag detection in 145 different altitudes, nor was it tested in different environments 146 with different tag densities. Having large populations of 147 RFID tags in an environment introduce more parameters that 148 may reduce the accuracy of the model. Cross interference is 149 another type of interference that can degrade the performance 150 of the simulator. It is most likely to occur between RFID 151 systems and WIFI or personal area networks (WPAN) such 152 as Bluetooth but only when devices share common or adja-153 cent frequency bands within the environment. In [20], authors 154 the knowledge of the theoretical maximum collision time 156 and collision probability between RFID and WiFi/Bluetooth 157 packets. This scheme generates an optimal channel based on 158 the current usage of the adjacent frequency channels thereby 159 reducing the interference. We conclude from the above, that 160 it is extremely complex to consider all the parameters that  or not, but to estimate how many tags will be read from a 210 given constellation of tags. When we compare with experi-211 ments, we will not compare tags on an individual basis, but 212 we will compare whether the simulation and the experiment 213 have read approximately the same number of tags, and at a 214 similar rate (tags read per second), which is the way accuracy 215 is calculated in RFID deployments.

216
The PD is defined for each tag-antenna pair as a function 217 of 6 arguments. The first 3 are the distance and two angle 218 coordinates of the tag with respect to antenna position and 219 its direction of maximum radiation: R, θ H , and θ V , which 220 in Gazebo are the representation of the transforms between 221 the sensor-frames [22] ''RFID-Antenna'' and ''RFID-Tag'', 222 which are automatically calculated by ROS.

223
The remaining 3 parameters are constants that depend on 224 the particular RFID system used and its RFID settings. These 225 parameters are R 0 , the distance at which half of the tags can 226 be read during a specific duration of time, and the antenna 227 beam widths in the horizontal and vertical planes, θ H and 228 θ V . R 0 depends mostly on the Equivalent Isotropic Radi-229 ated Power (EIRP) which is defined as the product of the 230 conducted power (P in ) and the antenna gain (G t ), EIRP = 231 P in · G t , as well as the sensitivity of the tags. θ H and θ V 232 depend on the particular reader antenna used in the system. 233 These 3 parameters must be supplied by the user of the 234 plugin. The antenna beam-widths are normally found in the 235 data sheet of the antenna used in the system, and R 0 must 236 be adjusted by calibrating the simulation against some 237 experiments.

238
R 0 is defined so that when R = R 0 and θ H = θ V = 0 the 239 probability of detection is PD = 0.5. At other distances 240 and angles, PD is calculated using the antenna pattern based 241 on the beam-widths, and the 1/R 2 decay of surface power 242 density.

243
Increasing R 0 (by increasing the EIRP, the gain of the 244 reader antenna, and/or the sensitivity of the tags), will allow 245 the antenna to detect distant tags with higher probability, 246 while decreasing it will tend to allow only the detection of 247 tags at shorter distances. On the other hand, using antennas 248 with wider beam-widths will allow the antenna to detect tags 249 at wider angles from the front direction of the antenna. An antenna's radiation pattern describes how the antenna 252 radiates/receives energy into/from all directions in space, and 253 is three-dimensional. In the model, we approximate the nor-254 malized (maximum value of 1) radiation pattern as the func-255 (1) 258 VOLUME 10, 2022 This approximation is valid for antennas with a 264 well-defined main lobe, and considers any radiation in the 265 back hemisphere as negligible.

266
Approximating the tag antenna as isotropic, and neglecting 267 any multipath interference, the received power by the antenna 268 is proportional to the reader antenna directivity and to 1 R 2 , 269 as shown in Eq. 2.
Given that the probability of detection We arbitrarily define it as: where x is always positive, and is defined as:  The pre-designed and patent UGV in [23] is used. This

341
UGV is used to carry the RFID payload, and its hardware 342 block diagram is shown in Fig. 4. The UGV is designed to (SLAM) based camera to supply the UAV with self-382 localization coordinates. Special adaptation was made 383 to infuse these coordinates to the autopilot, resulting 384 in an indoors guidance system for the UAV. This part 385 is responsible for adding intelligence to the contiguous 386 main flight system part. It also enables the possibility to 387 sense the environment and obstacles nearby through a 388 depth proximity camera. All these sensor data, includ-389 ing the data received from the RFID-Payload, are pro-390 cessed by a companion computer (CC). The CC will 391 incorporate the navigation algorithm. This same algo-392 rithm will be used for the simulation and laboratory 393 experiments. The output of this CC will be mainly the 394 control signals in the form of pose-goal, or movement 395 commands to the autopilot.

436
For all the Experiments in these Scenarios, both robots will 437 be tested with a constant velocity of 1m/s. However, lower 438 or higher velocities could be used. It is important, to take 439 into consideration the differences and the effects that the 440 actual physical environment has on both robots as compared 441 in simulation. These effects could be slight odometry or 442 wheel misplacement errors due to the ground surface for 443 a UGV, to small displacements in the UAV position, while 444 constantly trying to stabilize itself in the air, this is due

467
In Experiment 2, we repeat Experiment 1 but using a  the effect of the orientation of the RFID antenna on the total 481 number of unique RFID tags read is more relevant. For this 482 reason, we run Experiment 2 four times, and for each one 483 we activate only one of the four RFID antennas mounted on 484 the UAV with different orientations, as illustrated in Fig. 5    each with one of the four antennas active. We observe 512 in Figs. 18a, 18b, 18c, and 18d, that results are in good 513 agreement for all four antennas even though the tag den-514 sity was doubled. Figs. 17a, 17c, 17e, and 17g, illustrate the 515 position of the detected RFID tags in simulation, while 516 Figs. 17b, 17d, 17f, and 17h, illustrate the position of the 517 detected RFID tags in the laboratory, which are also in very 518 good agreement. 519 We can notice in Figs. 17g and 17h, that with only the left 520 antenna active on the UAV, the majority of the detected unique 521 RFID tags were located on the edges of the map environment. 522 On the other hand, in Figs. 17c, and 17d, having only the 523 right antenna active, we see that most of the detected tags 524   The ability for the RFID system plugin to simulate accu-547 rately the behavior of an RFID system payload (reader 548 and antennas) on board an operating mobile robot in an 549 environment where RFID tags are present, enables it to 550 be used in research that involves RFID and Robotics 551 VOLUME 10, 2022 technologies working together. For this application scenario, 552 we utilize the RFID system plugin to enable a UAV to navi-553 gate using stigmergy [28] to inventory a space using RFID 554 technology. In this scenario, we will have a UAV navigat-555 ing autonomously through an environment where RFID tags 556 are present, but whose quantity and position are unknown.  reads more unique tags. Research on finding new UAV nav-563 igation strategies is expensive in time and money, and the 564 possibility to run simulations to validate the algorithms can 565 speed up the process considerably. But when RFID payloads 566 are used, the simulation is only possible if a simulation tool 567 is available for the RFID system, such as the RFID sys-568 tem plugin that we propose. For this simulation 300 tags 569 were uniformly placed in a horizontal manner, throughout the 570 T-shaped layout. The simulation and laboratory setups are 571 shown in Figs. 22a and 22b. Fig. 22c shows the path of the 572 UAV in simulation, and Fig. 22d shows that the UAV was able 573 to read 283 tags out of 300 overall tags in the environment 574 page and repository [21].

578
In summary, this paper addresses the problem of simulating 579 RFID systems, including readers, antennas and tags, being