PARTICLE FILTER APPROACH TO UTILIZATION OF WIRELESS SIGNAL STRENGTH FOR MOBILE ROBOT LOCALIZATION IN INDOOR ENVIRONMENTS

Transcription

1 PARTICLE FILTER APPROACH TO UTILIZATION OF WIRELESS SIGNAL STRENGTH FOR MOBILE ROBOT LOCALIZATION IN INDOOR ENVIRONMENTS Samuel L. Shue 1, Nelyadi S. Shey 1, Aidan F. Browne 1 and James M. Conrad 1 1 The William Saes Lee College of Engineering, The Universiy of Norh Carolina a Charloe, Charloe, Norh Carolina, USA ABSTRACT For many auonomous roboic applicaions, he capabiliy o simulaneously creae a map of he environmen while localizing is posiion wihin i is of criical imporance. This is ypically achieved by fusing odomery informaion from he roboic vehicle wih informaion from landmarks deeced wihin he environmen. Indoor environmens ofen have exising wireless infrasrucure, which can be used as landmarks by esimaing he disance beween he robo and he access poin. The mos pracical way o aain his is by measuring he decay of signal srengh. However, radio signal srengh does no predicably aenuae indoors as i does in open environmens due o signal inerference, absorpion, and reflecion from objecs wihin he environmen, inflicing unexpeced amplificaion or decay a he receiver known as mulipah inerference. This causes erroneous disance esimaions due o he unexpeced changes in signal srengh aenuaion. In his research, models of radio propagaion as i relaes o he received signal srengh indicaor (RSSI) are explored along wih localizaion echniques which uilize hese models. For developmen and esing of RSSI-based localizaion echniques, a simulaion mehod has been described which uilizes a Markov chain o provide realisic mulipah inerference on simulaed RSSI daa. Using his simulaion echnique, mehods for simulaneous localizaion and mapping (SLAM) are explored. Due o he difficuly associaed wih modeling RSSI aenuaion and disance esimaion, a paricle filer based SLAM approach is proposed and demonsraed. KEYWORDS RSSI, ZigBee, Localizaion, Paricle Filer, SLAM & RO-SLAM 1. INTRODUCTION In recen years, mobile roboics has grown ino an enormous research field for boh indusrial and domesic applicaions. These applicaions include auonomous vacuum cleaners, pool cleaners, lawn mowers, and self-driving vehicles, jus o name a few. In all of hese applicaions, knowledge of he robo's posiion wihin is operaing environmen is of criical imporance. To accuraely fulfill he requiremens of any mobile roboics ask, he robo mus be aware of is posiion relaive o he objecs wihin he environmen o properly inerac wih hem. The robo mus also be able o build a map o assis in navigaing from poin o poin wihin ha area. This presens a causaliy dilemma, in which he robo mus know is own posiion o build a map, bu also know he map o rack is posiion. Algorihms ha aemp o solve his problem are known as Simulaneous Localizaion and Mapping (SLAM) algorihms. Robos use a variey of sensors o esimae heir posiion, such as odomery measuremens from he mobile base as well as range finding sensors or visual informaion from cameras o help DOI: /ijwmn

2 idenify recognizable landmarks wihin he environmen. In his research, wireless signal srengh is used as sensory informaion o esimae he disance beween ransceivers. If he robo is equipped wih a wireless ransmier, as he signal aenuaes while raveling o he receiver, disance o ha receiver can be esimaed based on he loss of signal srengh [1]. This loss of power beween ransceivers is represened by he Received Signal Srengh Indicaor (RSSI). Mos wireless ransceivers represen RSSI in -dbm (decibel milliwas) which represens he amoun of aenuaion or los power during ransmission. If he operaing environmen of a wireless mobile robo is equipped wih ransceivers scaered hroughou, he esimaed disance o each ransceiver from he robo can be used o refine he posiion esimaion of he robo [2, 3]. There are several mahemaical models which can be used for esimaing a disance from RSSI. These models largely do no accoun for he mulipah fading effec. Mulipah fading occurs when a signal akes muliple pahs from he ransmier o he receiver. Depending on he phase of signals when hey arrive ogeher, hey can eiher cause eiher large aenuaion or large amplificaion of he signal. When he signal is suddenly amplified, he model will no hold and will generae highly erroneous readings. Some models accoun for mulipah bu are inended for long-range ransmissions, such as ransmission from a cell ower [4, 5, 6]. Mulipah fading effecs in shor-range indoor environmens are much more difficul o predic due o he many facors ha affec is behavior. For mobile robo operaing environmens, such as a warehouse or a privae home, shor-range mulipah is he mos prevalen problem in RSSI disance esimaion. Many soluions exis o miigae he effecs of mulipah aenuaion. However, mos of hese rely on some modificaion o he wireless modulaion hardware. Common sraegies involve analyzing he signal before he demodulaion phase, aemping o idenify mulipah componens and removing hem, leaving he primary pah inac. Ulra-Wideband echniques uilize various frequencies for ransmiing he same message and observing how each frequency is affeced by he ransmission, allowing he deecion and compensaion for mulipah inerference. Oher echniques will uilize anenna arrays or direcional anennas o avoid mulipah inerference. While hese mehods have proven effecive, hey all require some specialy modificaion o he ransmission and recepion hardware or uilize non-sandard communicaion proocols. This limis he use of hese echniques in exising neworks and an increase in implemenaion coss due o hardware obscuriy. For his reason, i would be desirable o find a soluion ha uilizes he common hardware. Mos wireless devices provide he user wih RSSI values for each ransmission. If mulipah inerference could be accouned for in RSSI-based ranging echniques, localizaion could be accomplished in a more inexpensive and accessible manner [2] Objecive In his work, wireless ransceivers will be uilized as landmarks for SLAM algorihms wih a focus on indoor environmens. SLAM algorihms use he robo's odomery informaion along wih deeced landmarks wihin he environmen o esimae he posiion and orienaion of he robo as well as building a map using range daa ypically from a laser rangefinder or camera. To effecively correc odomery errors, SLAM algorihms ypically need o collec and mainain many landmarks. This ends o cause mos SLAM algorihms o scale poorly over ime by consuming exponenially more memory as new landmarks are appended. Anoher issue encounered hrough he use of landmarks is he daa associaion problem, where a newly observed landmark may be improperly idenified as a previously recorded, differen landmark. Uilizing wireless ransceivers as landmarks largely solve hese problems, since he number of ransceivers is finie, limiing he number of landmarks o he number of devices wihin he nework. Generally, here are no enough nework devices where he memory consumpion becomes a problem for mos modern compuers and embedded devices. The landmark associaion problem is enirely solved as each landmark has a nework address associaed wih i. However, 22

3 he mulipah problem encounered when using wireless beacons renders many common SLAM algorihms inappropriae, such as algorihms based on he Exended Kalman Filer (EKF), which requires measuremen noise o be well modeled by a zero-mean Gaussian disribuion Conribuion This paper proposes a paricle filering SLAM configuraion o properly handle he non-gaussian disribued daa provided by RSSI range esimaion. This mehod is validaed hrough a simulaion which uilizes a unique RSSI simulaion echnique o provide realisic mulipah inerference and noise disribuion. The resuls are hen conrased wih he sandard EKF rangeonly SLAM mehodology. 2. BACKGROUND 2.1. RSSI Ranging Disance esimaion hrough signal srengh decay is one of he mos pracical mehods as mos wireless hardware inerfaces provide RSSI as an accessible parameer. Muliple models exis o describe he signal aenuaion over disance, however, he mos commonly used model is he logdisance pah loss model. The log-disance pah loss model is a general propagaion model. I can be used in boh indoor and oudoor environmens. The log-disance pah loss model provides a logarihmic aenuaion model which has several parameers ha can be uned o make i fi nearly any environmen [7, 8]. The RSSI (in dbm) for his model is expressed as: RSSI = 10n log 10 d + A Where n is he pah-loss exponen, d is he ransmission disance in meers, and A is he reference value, which is he RSSI a 1 meer away from he ransmier. This equaion can be rearranged o be expressed as a disance for a given RSSI value: d = 10 RSSI A 10n The pah loss exponen, n, can be calculaed for each environmen by recording RSSI values a known disances and reverse solving for n. Typical values for n can be observed in Table I [8, 9, 10]. Table 2.1: Sample pah-loss exponen values for various environmens. The log-disance pah loss model is he mos versaile, as i can be configured for each environmen and uses a reference value raher han requiring he ransmission power and gain for each ransmier and receiver. 23

4 2.2. EKF SLAM The Exended Kalman Filer (EKF) SLAM is one of he mos common and reliable implemenaions of SLAM. I can be adaped for muliple ypes of landmarks, can be quickly compued, and has a relaively sraigh-forward implemenaion. As he name implies, i uses he EKF a is core for handling he uncerainies associaed wih SLAM. EKF is a recursive Bayesian filer which models each inpu and predicion wih zero mean Gaussian noise [10] Paricle Filer Paricles filers are a caegory of Mone Carlo algorihms ha are used o esimae saes in parially observable Markov chains [11]. Mone Carlo algorihms represen he disribuion of a random variable hrough a large number of samples. This non-parameric mehod of represening a disribuion is appropriae for siuaions such as RSSI-disance esimaion in an indoor environmen, where he disribuion of he noise is nearly impossible o model. The paricle filer represens he sae of he sysem hrough a daase, X, composed of M random samples known as paricles. Each paricle, x, represens he hypohesis of ha paricular sae variable a ime. Like he Kalman filer, he paricle filer can be represened wih wo basic seps: predicion and measuremen/re-sampling. The predicion phase uses a conrol inpu, u, o updaes each paricle s value, x [m], and poserior disribuion, p(z x ), based on sysem dynamics. The measuremen phase uses sensory informaion, z, o validae he predicion of each paricle. The error beween he measured sensory informaion and he prediced sae is compued for each paricle, and a weigh, w, is generaed based on ha error. The paricles in he filer are hen randomly sampled according o he disribuion formed by he weigh assigned o each paricle (paricles wih prediced saes ha closely mached he measured sae are assigned a higher weigh and are more likely o be seleced). The paricles are resampled M imes, once for each paricle in he filer. Over many ieraions, he paricles converge owards he rue sae of he sysem [11, 12]. The algorihm used for he Paricle Filer (X 1,z ) is as follows: 1: X =X =0 2: for m=1 o M do [m] [m] 3: sample x p(x u,x ) 4: w =p(z x ) [m] [m] -1 [m] [m] 5: X =X +<X,w 6: for m=1 o M do 7: draw i wih probabiliy w [i] [i] 8: add x o X 9: reurn X Paricle filers have been successfully able o solve applicaions in he roboics domain for more han one reason. They can be applied o any model ha is probabilisic and can be represened as a Markov chain. Secondly, paricle filers do no require a fixed compuaional ime raher heir accuracy increases wih increase in compuaional resources as menioned in [13] and [14]. 24

5 Figure 1: Convergence of paricle filer for mobile robo localizaion [11]. 3. METHODOLOGY 3.1. Implemenaion This research proposes a pure paricle filer RO SLAM implemenaion for RSSI ranging applicaion. This implemenaion describes each aspec of he SLAM algorihm wih individual paricle filers. Iniially, a paricle filer is defined for he locaion and orienaion of he mobile robo. As wih mos SLAM algorihms, he robo is defined iniially and he map is creaed wih he origin cenered on he robo s saring posiion. These paricles begin wih a high cerainy cenered on his locaion. As he robo receives conrol inpus, u, from he odomery he paricles and heir poseriors are compued. However, he resampling phase canno be compued unil landmarks are idenified wihin he environmen. When he robo receives a new RSSI reading from an uniniialized landmark, a new paricle filer is generaed for ha landmark. Unlike he robo, he landmarks have a highly uncerain posiion. Therefore, he paricles are iniialized randomly from a uniform disribuion across he enire region of ineres. The landmarks have no conrol inpu, bu slowly converge as measuremens are aken by he mobile robo. The expeced range value is compued for each paricle using he expeced value from he mobile robo paricle filer and he hypohesized locaion of he landmark for ha paricle. Once a landmark paricle filer has converged o an accepable level, i can be used o resample he mobile robo paricle filer. The robo s locaion paricles are resampled in a similar manner o he landmarks, using he expeced value of he landmark paricles o generae he expeced values for each paricle. This process is deailed in he following implemenaion algorihm: 25

6 1:for n = 1 o number of landmarks+1 do 2: X ( )( n) = paricles uniformly disribued 3: X ( )( n) =0 4: Measuremen Sep Landmarks: 5: for n = 1 o number of landmarks do 6: for = 1 o number of paricles do 7: [ n ] [n] = ( z X ) w p [n] [n] [n] [n] 8: X = X + < X,w 9: for = 1 o T do 10: draw i wih probabiliy 11: [i] [n] add x o X [1:N] 12: reurn X 13: Measuremen Sep Robo: 14: for n = 1 o number of landmarks do 15: for = 1 o number of paricles do 16: w =p (z X ) [n] [N+1] [N+1] [N+1] [N+1] [n] 17: X = X + < X,w 18: for = 1 o T do 19: draw i wih probabiliy w [i] 20: [i] add x o X [N+1] 21: reurn X 22: Predicion Sep: 23: for = 1 o number of paricles do [N+1] [N+1] 24: x» p( x u,x -1 ) As wih all RO SLAM algorihms, a landmark mus go hrough an iniializaion phase prior o being used for measuremen. This is accomplished in his implemenaion by waiing unil he paricles for a landmark have converged o an accepable level. However, when using RSSI ranging, i is possible for a landmark o converge while undergoing mulipah inerference, creaing a highly erroneous iniializaion. When his occurs in EKF based algorihms, i can be difficul o recover he landmark, as he Gaussian assumpion prevens he esimaion from rapidly updaing when leaving he area undergoing he mulipah inerference. The paricle filer s adapive naure makes his scenario less impacful on he overall sae of he filer. Figure 2: Implemenaion of Paricle Filer used o perform SLAM. 26

7 The red dos and red circle are he acual landmark and robo posiions. The blue, purple, yellow, orange paricles represen each landmark. The green se of paricles represen he robo paricle filer and he black circle represens he expeced value of he robo pose. 4. RSSI SIMULATION METHOD An RSSI simulaion mehod was used o creae he environmen used o evaluae he mehod described in he previous secion. This simulaion mehod uilizes a Markov chain rained on colleced RSSI daa o produce an RSSI-disance daase. The Markov chain generaes a lookup able ha is used o access an RSSI value when he robo is a a specific posiion relaive o he wireless node. This simulaion mehod capures he complex effecs of mulipah inerference wihou he need he model he environmen, which is usually a edious and compuaionally expensive process as mulipah is deermined by no only he geomery of he environmen bu also he refleciveness of he maerials which compose i. This simulaion is described in deail in [15]. For he examples shown in his work, he Markov chain was rained on daa colleced wihin he Energy Producion and Infrasrucure Cener (EPIC) building on he campus of he Universiy of Norh Carolina a Charloe. Daa were colleced in various environmen ypes, including large classrooms, hallways, and laboraories. An oudoor daase was also colleced for a se wih minimal mulipah inerference. This daa was colleced using XBee radios and Digi 2.4 GHz Omnidirecional Dipole anennas wih a 2.1 dbi Gain. The XBee radios were configured wih boos mode disabled and power level se o 2 (1 dbm Gain). Each XBee module was inerfaced wih an Arduino Uno; one module being se o ransmi, and he oher configured o reurn he RSSI value of he received packe. Each node was placed on a 1 meer PVC pole and displaced from he ransmier a 0.25-meer incremens up o 10 meers. A each incremen, 5 readings were aken a 90-degree incremens for a full roaion, o accoun for any asymmeric radiaion paerns of he anennae and is effec on he RSSI reading [15, 16]. 5. RESULTS Using he RSSI simulaion mehod described in 3.2, simulaion environmens were creaed o validae he proposed paricle filer mehod. Differen es cases were simulaed for various RSSI daases used. The es cases for each environmen are furher divided ino landmarks iniialized a rue locaions and falsely/erroneously iniialized landmarks. Boh cases use he same se of RSSI signals generaed by our simulaor. The firs graph in each case represens he disribuion of he noise wih mulipah from he radio beacons. This is obained hrough he difference beween he acual robo and landmark o he disance derived from he signal generaed by our simulaor. The following graphs represen he accumulaed robo posiion error for he differen mehods, landmark localizaion error and he acual pah followed by ground ruh, dead reckoning, EKF RO SLAM, and he proposed Paricle Filer RO SLAM. In each case, he robo is driven wih a consan linear velociy of 0.1 m/sec in an S-Shaped rack and an angular velociy of 30 degrees a he urns. To save compuaion ime, i is assumed he landmarks have already been iniialized/converged wih some amoun of error. To ensure each paricle filer used in he proposed mehod represens he disribuion well, 1000 paricles are generaed for each filer in hese simulaions. 27

8 5.1. Large Open Room Environmen Figure 3: Error in disance derived from RSSI(in meers) in a large open room environmen. Figure 3 displays he hisogram of he beacon noise which has a variance of 3.5 meers. This value is assigned as he variance of he measuremen noise in RO EKF. Also, a Gaussian zero mean odomery noise wih a variance of 0.1 meers linear velociy is inroduced, a zero mean Gaussian noise of 10 degrees is used for he angular velociy noise. I uses a paricle filer ha mainains 1000 paricles for each landmark and robo. True Landmark Iniializaion Figure 4: Error in he pah followed in a large open room environmen. Figure 5: Error in posiion accumulaed over he pah in a large open environmen over ime. 28

9 Figure 6: Error in he landmark localizaion in a large open room environmen over ime. In his simulaion, he mulipah inerference is minimal for an indoor environmen, which is refleced in Figure 5 where he locaion error beween he paricle filer mehod and RO EKF are similar and low. This is due o he fac he noise in he large open room has a beer Gaussian approximaion han wha was observed in he following experimens. Wih minimal mulipah inerference on he signal, i is expeced for RO-EKF and he paricle filer o exhibi comparable performance False Landmark Iniializaion Figure 7: Error in he pah followed in a large open room environmen wih false iniializaion. Figure 8: Error in he robo s posiion in a large open room environmen wih false iniializaion over ime. 29

10 Figure 9: Error in he landmark localizaion in a large open room environmen wih false iniializaion over ime. In his experimen, he landmarks are given incorrec iniial condiions when added o each SLAM algorihm, as shown in Figure 7. Despie he RSSI noise being near-gaussian, he approximaion does no serve well when aemping o correc fauly landmark iniializaion. The fauly landmarks and inappropriae Gaussian noise represenaion causes he false landmarks o generae fauly correcions o he robo s pose, causing he filer o become more erroneous over ime Hallway Environmen Figure 10: The hisogram of he error in RSSI signals (in meers) generaed by he simulaor by using he daase represening hallway environmens. In his case, a measuremen noise of 3.25 meers variance is applied o RO EKF SLAM measuremen noise and repeaed he process menioned in he previous es case. 30

11 True Landmark Iniializaion: Figure 11: Error in he pah followed in Hallway environmens. Figure 12: Error in he robo s posiion accumulaed over he enire pah in a Hallway environmen. Figure 13: Error in he landmark localizaion in a Hallway environmen. 31

12 The hallway environmen has a poorly approximaed Gaussian noise disribuion, which creaes huge problems for he RO-EKF mehod. Here, he paricle filer mehod is able o adap o he shifs in noise disribuion from mulipah inerference, keeping he localizaion error below ~1 meer, which is o be expeced for RSSI applicaions False Landmark Iniializaion Figure 14: Error in he pah followed in a Hallway environmen wih false iniializaions. Figure 15: Error in he robo s posiion in a Hallway environmen wih false iniializaion. Figure 16: Error in he landmark localizaion in a Hallway environmen wih false iniializaion. The hallway experimen wih fauly landmark iniializaion proves even more roublesome for RO-EKF, causing erroneous measuremens and noise represenaion o accrue rapidly. The paricle filer robo pose locaion error remains similar o he previous experimen, however, i is also able o correc he beacon locaions as well. 32

13 5.3. Furnished laboraory environmen Figure 17: The hisogram of he error in RSSI signals (in meers) generaed by he simulaor by using he daase represening a lab. In his case, we have used a measuremen noise of 2 meers variance in our RO EKF and have repeaed he process menioned in he previous es case True Landmark Iniializaion Figure 18: Error in he pah followed in a lab environmen. Figure 19: Error in he robo s posiion accumulaed in a Lab environmen over ime. 33

14 Figure 20: Error in he landmark localizaion in a Lab environmen over ime. The furnished laboraory environmen proves he mos difficul of all due o he many reflecive, meal desks, drawers, and chairs scaered abou he room. The disribuion in Figure 17 is mulimodal, causing RO-EKF o perform poorly despie correc landmark iniializaion. The paricle filer mainains similar accuracy o he oher environmens here as well False Landmark Iniializaion Figure 21: Error in he pah followed in lab environmens wih false iniializaion. Figure 22: Error in he robo s posiion in a Lab environmen wih false iniializaion over ime. 34

15 Figure 23: Error in he landmark localizaion in a lab environmen over ime. Finally, he laboraory environmen wih incorrec landmark iniializaion creaes he mos nonideal RO-EKF environmen, causing EKF o perform similarly o he odomery informaion. This is expeced as he variance is so high compared o he variance of he odomery, he correcions made from he RSSI measuremens are minimal. Once again, he paricle filer performs as well as i does in all oher condiions, along wih correcing he landmark posiions Comparison of Environmens As shown in Table 4.1, he EKF becomes almos comparable o he paricle filer when he landmarks are correcly iniialized depending on he amoun of mulipah i comes across as i raverses hrough he environmen. Bu when falsely iniialized i becomes comparable o dead reckoning. Table 4.1 conains he oal error for each of hese experimens for easy comparison. Table 4.1: Comparison of accumulaed robo locaion error over he enire pah using paricle filer, dead reckoning and EKF mehods in differen environmens Comparison regarding he number of paricles used. Below we have presened he differences in he efficiency in using differen numbers of paricles. We ran he proposed paricle filer mehod, dead reckoning, and Range Only EKF for a lab environmen. We repeaed he process for he paricle filer using 2000, 1000, 500, 300 and 150 paricles. The cases were run using he same daase generaed by he RSSI signal simulaor. 35

16 Figure 24: Paricle filer convergence ime for various number of paricles. As seen in Figure 24, as we reduce he number of paricles we observe a rade-off in he convergence ime and error incurred over he pah wih he paricle filer wih 150 paricles aking he longes ime o converge along wih he mos accumulaed error incurred. 6. CONCLUSIONS Here mehods for RO SLAM using RSSI ranging have been explored wih a focus on operaion wihin indoor environmens. These mehods have a high probabiliy of failure in highly noisy environmens due o he effec mulipah propagaion has on range esimaion from RSSI daa. This causes radiional EKF RO SLAM algorihms o diverge and hence resuling in false resuls because mulipah creaes highly non-zero, non-gaussian signal noise which violaes he requiremens for proper EKF funcionaliy. Here, a paricle filer based RO SLAM approach is presened o accoun for his issue, as paricle filer approaches are appropriae for scenarios in which he noise disribuion is no easily modelled. A novel configuraion for he paricle filer is proposed and implemened in a simulaion using realisic RSSI measuremens. I is compared o RO EKF SLAM and dead reckoning in differen environmens varying in he degrees of mulipah propagaion. I is esed for boh rue and false landmark iniializaions. Performance merics such as oal error incurred over he pah and landmark localizaion error is colleced and compared. The noise from he beacons is also compared o show he disribuion in he error. In conclusion, wih rue landmark iniializaion, he EKF and he paricle filer incurs comparable amouns of error depending on he mulipah signals received. However, he EKF copes poorly wih he non-gaussian noise presened by mulipah. When i is falsely iniialized in RO EKF i barely updaes he robo s posiion from he measuremen and hence i accumulaes he error incurred from odomery and becomes comparable o dead reckoning. The non-gaussian disribuion of he measuremen noise furher causes he landmark o be inaccuraely correced. The resuls also proposed o have he minimum number of paricles required o represen he error disribuion in he RSSI signals so ha posiions relaed o mulipah propagaion have some probabiliy of being re-sampled along wih he line of sigh signals. In his case, anyhing greaer han 1000 paricles has negligible improvemen in resuls. Furher research includes he implemenaion and evaluaion of he proposed paricle filer SLAM mehod on a physical mobile robo plaform and using real- ime sensor measuremens. While he effec of he number of paricles needed for proper disribuion represenaion is explored here, his may no hold rue for oher indoor environmens no evaluaed in his work and he error induced by non-simulaed odemeric error. Evaluaion of his mehod in oher mediums is of ineres as well, apar from radio signal propagaion. RO-SLAM is required for underwaer 36

17 auonomous vehicle applicaions where ulrasonic beacons mus be used for landmarks in areas where he sea floor does no provide disinguishable landmarks. The mulipah effec is sill presen in ulrasonic beacons and odomery error may even be more non-gaussian in he face of oceanic currens. REFERENCES [1] M. L. Rodrigues, L. F. M. Vieira and M. F. M. Campos, "Mobile Robo Localizaion in Indoor Environmens Using Muliple Wireless Technologies," 2012 Brazilian Roboics Symposium and Lain American Roboics Symposium, Foraleza, 2012, pp [2] S. L. Shue, Uilizaion of wireless signal srengh for mobile robo localizaion in indoor environmens (Docoral disseraion). The Universiy of Norh Carolina a Charloe, ProQues Disseraions Publishing, [3] Z. Xu, K. Sandrasegaran, X. Kong, X. Zhu, J. Zhao, B. Hu, C.-C. Lin, "Pedesrain monioring sysem using Wi-Fi echnology and RSSI based localizaion", Inernaional Journal of Wireless & Mobile Neworks (IJWMN), vol. 5, no. 4, Aug [4] S. Sain, Modelling and Characerizaion of Wireless Channels in Harsh Environmens, Thesis, Mälardalen Universiy, no. 1154, pp. 136, [5] Two-ray ground-reflecion model - Wikipedia. hps://en.wikipedia.org/wiki/two-ray_groundreflecion_model. (Accessed on 03/28/2017). [6] J. S. Seybold, Inroducion o RF Propagaion. John Wiley and Sons, [7] J. Xu, W. Liu, F. Lang, Y. Zhang, C. Wang, "Disance Measuremen Model Based on RSSI in WSN," Wireless Sensor Nework, Vol. 2 No. 8, 2010, pp [8] S. Shue, L. E. Johnson and J. M. Conrad, "Uilizaion of XBee ZigBee modules and MATLAB for RSSI localizaion applicaions," SouheasCon 2017, Charloe, NC, 2017, pp [9] S. Elango; N. Mahivanan, and G. Pankaj. RSSI Based Indoor Posiion Monioring Using WSN in a Home Auomaion Applicaion. Microprocessors and Microsysems. 2011; 11(4): [10] R. Mehra; A. Singh, "Real ime RSSI error reducion in disance esimaion using RLS algorihm," in Advance Compuing Conference (IACC), 2013 IEEE 3rd Inernaional, pp , Feb [11] S. Thrun, Probabilisic roboics, Communicaions of he ACM, vol. 45, no. 3, [12] S. Thrun, Paricle Filers in Roboics, Proceedings of Uncerainy in AI, vol. 1, pp , [13] G. Grisei, C. Sachniss, and W. Burgard, Improved Techniques for Grid Mapping Wih Rao- Blackwellized Paricle Filers, IEEE Transacions on Roboics, vol. 23, pp , Feb [14] G. Grisei, R. Kummerle, C. Sachniss and W. Burgard, "A Tuorial on Graph-Based SLAM," in IEEE Inelligen Transporaion Sysems Magazine, vol. 2, no. 4, pp , winer [15] S. Shue and J. M. Conrad, Procedurally generaed environmens for simulaing RSSI- localizaion applicaions, in Proceedings of he 20h Communicaions & Neworking Symposium, CNS 17, (San Diego, CA, USA), pp. 7:1 7:11, Sociey for Compuer Simulaion Inernaional, [16] Nelyadi Samyak Shey, Paricle Filer Approach o Overcome Mulipah Propagaion Error in Slam Indoor Applicaions (Maser s hesis). The Universiy of Norh Carolina a Charloe, ProQues Disseraions Publishing,

18 Auhors Inernaional Journal of Wireless & Mobile Neworks (IJWMN) Vol. 10, No. 4, Augus 2018 SAMUEL L. SHUE (M 14 PhD 17) received his B.S. in compuer engineering in 2011, M.S. in elecrical engineering in 2015, and his Ph.D. in elecrical engineering in 2017 from he Universiy of Norh Carolina a Charloe. He currenly works as an embedded sysems engineer for NLA Diagnosics in Charloe and is also an adjus professor a UNCC. He is also an acive member of IEEE and he IEEE Roboics and Auomaion Sociey. His research ineress include localizaion and mapping for roboics, wireless localizaion, signal processing, and embedded sysems. NELYADI SAMYAK SHETTY (M 18) received he B.S degree in elecronics and communicaion engineering from he Bangalore Insiue of Technology, Bangalore, India in 2015 and he M.Sc. degree in elecrical engineering from he Universiy Of Norh Carolina Charloe, Norh Carolina in 2018, where he worked on developing robus algorihms for robo localizaion using radio signals in environmens ha exhibi high mulipah noise. His research ineress include embedded sysems, digial signal processing and Simulaneous Localizaion and Mapping algorihms for roboics applicaions. AIDAN F. BROWNE (M 95, 97 PhD 98) received his B.S. degree in physics and mahemaics from Vanderbil Universiy, Nashville, TN in 2008, an M.S. degree in biological engineering, an M.S. degree in elecrical engineering, and a Ph.D. in biomedical engineering, from he Universiy of Connecicu, Sorrs, CT, in 1995, 1997 and 1998, respecively. He is currenly an Assisan Professor in he Deparmen of Engineering Technology and Consrucion Managemen a The Universiy of Norh Carolina a Charloe. His research ineress include roboics, mecharonics, insrumenaion and sensing. Dr. Browne currenly serves as he Chair of he IEEE Charloe Secion. He holds memberships in he American Sociey of Engineering Educaion (ASEE) as well as he Vibraion Insiue. JAMES M. CONRAD (M 87 PhD 92) received his BS degree in compuer science from he Universiy of Illinois, Urbana, in 1984 and his M.S. and Ph.D. degrees in compuer engineering from Norh Carolina Sae Universiy in 1987 and 1992, respecively. Since 2003 he has been an Associae Professor and Professor wih he Elecrical and Compuer Engineering Deparmen a The Universiy of Norh Carolina a Charloe. He is he auhor of eigh books and more han 150 aricles in he areas of embedded sysems, roboics, parallel processing, and engineering educaion. Dr. Conrad was eleced o serve on he IEEE Board of Direcors as Region 3 Direcor/Delegae for He has been recognized by IEEE Region 3 and IEEE-USA wih several awards for exemplary eaching and service. 38