Experimental Validation of a MATLAB Based Control Architecture for Multiple Robot Outdoor Navigation

Transcription

1 Proceedngs of the 13th Medterranean Conference on Control and Automaton Lmassol, Cyprus, June 27-29, 2005 Expermental Valdaton of a MATLAB Based Control Archtecture for Multple Robot Outdoor Navgaton WeP04-2 L. Dotsds, A. L. Nelson, K. P. Valavans, M. T. Long, R. R. Murphy Abstract Desgn, mplementaton and expermental valdaton of a MATLAB based autonomous robot control framework s presented, supported by, and ntegrated nto a dstrbuted feld robot archtecture known as dstrbuted-sfx. The MATLAB based framework s comd of mult sensor fuzzy logc robot controllers that utlze laser, GPS and odometer data, fusng such sensor data and flterng out nose, to mprove collson free navgaton. Extensve outdoor envronment experments wth sngle and multple moble robots are performed to demonstrate waypont and goal pont navgaton, and raster scan search patterns n unknown envronments wth statc and dynamc obstacles. Results and vdeos are provded to justfy the prod approach. I. INTRODUCTION A dstrbuted robot control archtecture, both durng ts desgn as well as durng ts mplementaton and testng phase, brngs together concepts and deas -under a common framework- from the dverse felds of Dstrbuted Artfcal Intellgence, Human Robot Interacton and Mult Agent Systems (hgher levels/layers), coupled wth Control Theoretc approaches (lower level real-tme control), operatng n unson to provde heterogeneous system functonalty, adaptablty and flexblty. Ths paper focuses manly on the mplementaton and expermental valdaton of the control theoretc MATLAB based lower level of the archtecture, hghlghtng brefly for completeness purs ts ntegraton wth the overall system archtecture, referred to from now on as dstrbuted- SFX archtecture. Desgn and mplementaton detals of the MATLAB based framework are dscussed; mult sensor fuzzy logc robot controllers are derved and mplemented usng laser, GPS and odometer data as nputs; detals related to fusng such sensor data (laser, GPS and Ths work has been partally supported by a grant form ONR, N (wth USF LO). L. Dotsds s supported partally by HERAKLITOS fellowshps for research from the Techncal Unversty of Crete, EPEAEK II L. Dotsds was wth Unversty of South Florda. He s now wth Department of Engneerng and Management, Techncal Unversty of Crete, Greece (e-mal: ldotsds@ depm.tuc.gr). A. L. Nelson, was wth the Center for Robot Asssted Search and Rescue, Unversty of South Florda. K. P. Valavans, M. T. Long, R. R. Murphy are wth the Center for Robot Asssted Search and Rescue, Unversty of South Florda. odometer) flterng out nose, are presented; sngle and multple robot waypont navgaton, goal pont navgaton, random and raster scan search patterns along wth collson avodance are also demonstrated expermentally. A major novel aspect and contrbuton of ths research s related to the fact that although the dstrbuted-sfx archtecture s mplemented n Java and uses Jn to manage dstrbuted objects and servces between robots and computers [39], the MATLAB envronment s supported wthn the Java / Jn framework usng the JMatLnk lbrary [30] allowng for mathematcal control theoretc research and expermentaton and for rapd prototypng of behavoral and control modules and servces. Wrappng the MATLAB workspace envronment wth JMatLnk, n conjuncton wth the Jn dstrbuted object platform, allows modules and servces mplemented as natve nterpreted MATLAB code to be accessed as remote and dstrbuted objects. Although t s the case that many behavoral archtectures and languages support mplementaton of modules n dfferent languages, ths approach allows MATLAB to be drectly ncorporated nto behavoral archtectures resultng n speed up of development and added flexblty of mplementaton. The combnaton of Java based dstrbuted archtectures usng MATLAB n ts nterpreted form for autonomous robot navgaton and control s not well represented n the lterature, and up to now, no complete approach has been publshed. The derved fuzzy logc robot controllers utlze three dfferent sensors (lasers, GPS and odometers), they are capable of movng to one or more goal postons regardless of ntal poston or unpredcted devatons and obstacles along the paths traversed; they may be dynamcally redrected to reach a new set of goal or wayponts regardless of the robot current state and locaton; the robots move contnuously wthout stoppng to evaluate terran or process sensor data and navgate to goal ponts. The presented approach dffers consderably from related work n [18] and [19]; the reported research n [18] uses dead reckonng requrng a-pror knowledge of ntal poston, t s not robust aganst cumulatve odometer errors, and at tmes the robot remans statonary durng executon. The approach followed n [19] for outdoor navgaton uses small lab-based robots to prototype controllers for large /05/$ IEEE 1499

2 outdoor vehcles n agrcultural envronments; ndoor robots use herarchcal fuzzy controllers havng a set of dfferent behavors ncludng obstacle avodance, goal seekng, and edge followng, whch are then transferred to outdoor vehcles; however t s appled manly to corrdor and edge followng usng an IR beacon for homng on goal postons. Further, compared wth reported research n [5], [6], [7], the fuzzy controllers ntroduced here use GPS, odometer and laser data and they are appled n outdoor envronments, as opd to usng only sonar sensor data for ndoor navgaton. The expermental testbed conssts of a fleet of custom made ATRV-Jr skd steerng dfferental drve robots equpped wth 1 GHz P III processors and 2 Gb of RAM. On board computer ncludes 10 auxlary seral ports, frewre and UBS to accommodate dverse sensors and accessores; robots run RedHat Lnux 9.0. They are capable of runnng locally any standard PC software and they are accessble va wreless Ethernet (802.11g). The sensor sute ncludes a SICK LMS 200 scannng planar laser mounted on the front of each vehcle, a pan/tlt unt wth FLIR and standard vdeo cameras attached to the front-top equpment rack, IMU and GPS mounted on the center-top equpment rack. The GPS system s a Synergy Systems LLC M12+ wth evaluaton board and a HAWK GPS antenna connected va seral lnk to the robot s man computer. The robots are powered from 2 lead-acd 720 Watt/hour batteres and can be operated for two to three hours contnuously. A custom made 24V power supply box has been desgned and nstalled n each robot to provde power to added devces. Lookng at the lterature, there exst several moble robot control archtectures n the lterature, ncludng behavoral robotcs archtectures [2], [3], [4], [10], [11], [13], [20], [31], [34], [38], [39], desgned to support robot control wthn methodologcal and conceptual constrants as well as wthn the behavor-based paradgm. Utlzaton of MATLAB as the prmary computng envronment for robot control expermentaton allowng access to all sensors and actuators n the form of functons that appear as standard MATLAB functon calls has been reported n [32], [33]. Integraton of MATLAB and Java n engneerng systems s the topc addressed n [35], [38]. Handlng uncertanty n autonomous robot control s the topc of [1], whle methods employng fuzzy logc moble robot navgaton (mostly ndoors) nclude [5], [7], [10], [11], [12], [13], [14], [15], [16], [17], [18] and [19]. Waypont navgaton forcng vehcles to move n predefned envronments s reported n [26] and [27]. Outdoor envronment navgaton usng odometer data only has been proven nadequate due to sgnfcant cumulatve odometer errors [21], [22]; the use of GPS or even other sensors lke INS and gyro s essental as shown n [23], [24], [25]. Secton 2 presents essental nformaton on the dstrbuted SFX archtecture and ts nterface wth MATLAB, whle Secton 3 dscusses sensor data processng and robot poston detecton. Secton 4 presents the fuzzy logc robot control approach. Expermental results are the topc of Secton 5. Secton 6 concludes the paper. II. DISTRIBUTED SFX AND MATLAB The Dstrbuted-SFX archtecture [37] s a servce-based dstrbuted robot archtecture desgned to support behavor based control. It s a dstrbuted Java/Jn based descendant of the Sensor Fuson Effects archtecture (SFX) [34]. The system makes use of modules to mplement robot sensng and control related servces. The archtecture also supples hgh-level servce managers for the coordnaton and functonal ntegraton of low level modular servces. Modules are exported to a dstrbuted run-tme system as servces wth certan attrbutes and types. Servce may then be searched for (usng a dstrbuted-object lookup servce) based on functonal attrbutes rather than detals of actual mplementaton or physcal locaton. Ths type of archtecture allows a decouplng of clent and server. Clents, n an abstract sense, are nterested n servces wth partcular attrbutes that provde or consume certan types of data. The JMatLnk class lbrary [30] s used to connect Java to MATLAB. JMatLnk ncludes methods and objects that allow Java to ntalze a workspace, wrte data members of any format to the workspace, read from the work space, and execute command lne functons. MATLAB scrpts and functons may run locally on the robots as nterpreted code wthout the need to be compled nto stand-alone executables. MATLAB s supported at the drver module mplementaton level and t may be used as the natve server mplementaton of a servce. Wthn Java and Jn, the assocated server and proxy handle the remote overhead and nteracton wth other servces [29]. III. SENSOR DATA PROCESSING Sensor data are pre-processed, fltered and fused. In partcular, data from the laser scanner are fused wth odometer data and are fltered to reduce nose and ghost readngs caused by bouncng of the laser unt, by varatons n grass and vegetaton as well as other unforeseen outdoor envronment condtons. Integraton of past GPS sensor readngs fused wth odometer readngs results n smoothng of the GPS errors provdng more accurate robot poston detecton; ths s accomplshed wth the ad of a smple heurstc desgn flter. Ths secton presents detals for the laser scan flterng and poston detecton, both essental to desgnng and mplementng n real-tme the robot fuzzy logc controllers. A. Laser Scan Flterng The range of the laser scan s 180 degrees, centered on the 1500

3 robot body-attached reference frame n whch each consecutve pont s offset by 1 degree (181 total ponts per scan). Laser data s used to dentfy and avod statc and dynamc obstacles durng navgaton. Wthout loss of generalty, for all practcal purs offerng a trade off between computatonal complexty and real-tme control update cycle, the effectve laser scan area s dvded n three radal sectors labelled as Left Area, Center Area, Rght Area, denoted by W =1, 2, 3, each one ncludng further dvson n Close, Medum and Far regons. as shown n Fg. 1. Ths dvson s also used to derve the fuzzy logc robot controllers (dscussed n the next Secton) and to calculate the correspondng collson possbltes reflectng the presence / absence of obstacles n the robot path. Laser effectve range s expermentally verfed to be about 8 meters (25 feet). The left and rght areas have a wdth of 70 0 each and the center area of Center Area Amatrx R of transformed scans represents the last n laser scans wth all range readngs transformed nto the current robot reference frame can then be constructed: r( k) rt ( k 1) R = rt ( k n) Hence, objects detected n multple current and past scans appear as range readngs of smlar values n multple rows of R. Past Laser Scan at tme k-n α Current Laser Scan at tme k Left Area Far Rght Area l Object n Envronment Medum Current Poston of Robot Past Poston of Robot 8m o 70 o 40 o 70 Close Fg. 1: Laser scan area radal sectors dvded nto close, medum and far areas The laser scan flterng process ntegrates nformaton from the most recent scan at tme k, wth nformaton from up to n prevous scans, k-1, k-2,, k-n. Fg. 2.shows two laser scans, one taken at tme k (the most recent one) and another taken at tme k-n (a prevous one), wth α and l the relatve angular and lnear offsets of the scans at tme k and tme k-n, respectvely. Each laser scan s represented as a vector r of 181 ponts n polar form wth elements r, {1,181} representng the Eucldan dstance from the orgn of the robot bodyattached reference frame, wth angles mplctly defned by poston wthn r: r = r r [ ] 1 2 r181 To ntegrate nformaton from multple scans, each of the (prevous) past n {1, N}scans s transformed nto the reference coordnate frame of the most recent scan r(k) usng a standard rotaton and translaton transformaton T, calculated from the tme hstory of robot wheel odometer readngs: r T ( k n) = r( k n) T 8 m Fg. 2: Laser scans taken at tme k (most recent scan) and at tme k-n (a prevous scan). A statstcal flter s derved consderng the varance, the mean value and the maxmum range readngs of the elements of each of the columns of R. These are gven by the vectors of r var r µ,and r max respectvely, each of whch s of length 181. The fnal fltered laser vector s gven by: = r r r Φ r [ ] 1 2 r181 r, µ f r,var < v = r,max otherwse r Φ s the fnal fltered set of range readngs spannng the forward feld of the robot n ts current poston, and r,µ, r,var and r,max are the th elements of r µ, r var and r max, respectvely, and ν an approprately defned threshold value. Elements of scans pror to current tme k outsde the robot s current forward facng feld of vew (after transformaton) are dscarded n the above calculatons. B. Poston Detecton The robot p(k) at tme k s represented as a vector of three elements: p ( k) = [ x ( k) y ( k) θ ( k)] x and y are rectangular coordnates gven n lattude and longtude n the world coordnate frame, and θ s the robot s angular headng wth East defned as θ=0 and 1501

4 postve angles measured counter clockwse; k and k-1 represent current and mmedately prevous tme steps. The detecton system reles prmarly on GPS readngs for both locaton and orentaton calculaton, usng odometer readngs for smoothng. The GPS poston p gps (k) rectangular coordnates (x gps and y gps ) are obtaned drectly from the most recent GPS sensor readngs (usng the Motorola propretary flter suppled wth the hardware unt), whle the headng s obtaned dfferentally usng the current coordnate and a prevous coordnate. The current robot poston based on odometer readngs relatve to the last known real world poston s represented by p odo (k). A heurstc algorthm s derved that uses the dfference (or error) between p gps and p odo when selectng the next poston value. Ths error term s: perror = p gps p The algorthm consders an addtonal aggregate error term e agg when selectng the next poston value correspondng to the cumulatve worst case aggregate odometer error that may have accumulated over a sequence of controller tme steps. The ratonale of the algorthm s that f GPS and odometer readngs agree wthn a nomnal error threshold, the robot at tme k s based on GPS; otherwse s based ncrementally on odometer readngs relatve to the last known accurate. However, f the potental worst case ncremental error emax s exceeded, the system s forced to revert to GPS for a tme step to mantan localzaton wthn the world coordnate frame. Ths smoothes the GPS nose, but keeps the odometer based path from devatng by more than emax from the assumed to be poston of the robot. The algorthm s mplemented as follows: odo If e agg > e max p (k) = p gps (k) e agg = 0 Else f p error < e agg p (k) = p gps (k) e agg = 0 Else p (k) = p odo (k) e agg = + e agg IV. FUZZY LOGIC CONTROLLER Gven the clarfcatons for sensor data processng and poston detecton, the overall structure of the mult sensor control system s shown n Fgure 4. It conssts of four modules: the laser range flter, poston detecton, headng error calculaton and the actual fuzzy logc robot controller. The control system receves as nputs laser, odometer and GPS data, as well as a control reference nput (next waypont or goal pont). It outputs actuator commands n terms of robot rotatonal and translatonal velocty. The fuzzy logc controller s mplemented as a Mamdantype controller smlar to prevous work [5], [7]. The fuzzy controller nput from the fltered laser range block conssts of a three value vector wth components related to the dstance of the closest object n the left sector of the scan, n the center sector and n the rght sector, respectvely. Ths nformaton s used to calculate three collson possbltes left, center, rght reflectng potental statc / dynamc obstacles n the robot feld of vew, smlar to the approach followed n [5], [7], but for outdoor envronments. The fourth nput to the fuzzy logc controller s the robot s headng error calculated from the robot s current headng and the desred headng. Implementaton wse, each of the three aggregate range nputs ncludes three trapezodal membershp functons namely, close, medum and far. The nput lngustc varables are denoted as left dstance, rght dstance and center dstance correspondng the left area, rght area and center area sectors. The headng error nput uses four trapezodal membershp functons and one trangular membershp functon. Chosen membershp functons for the nput varables are emprcally derved based on extensve tests and experments. The value of each dstance nput varable d (correspondng to left area, center area, rght area) s fuzzfed and expressed by the fuzzy sets C,MD,A referrng to close, medum, and far as shown n Fgure 2. The range of the membershp functons for each d s between 0-8 meters. The value of the nput varable headng error, he, s fuzzfed and expressed by the fuzzy sets FL, L, AH, R, FR, referrng to far left, left, ahead, rght, and far rght, respectvely. The range of the membershp functons for the headng error, s between -180 and 180 degrees. The fuzzy logc controller has two output varables, translatonal velocty (tr) mplemented wth two trapezodal and one trangular membershp functons, and rotatonal velocty (rv) mplemented wth four trapezodal membershp functons and one trangular membershp functon The value of the output varable tr s expressed by the fuzzy sets ST, SL, F referrng to stop, slow, and fast. The value of the output varable rv s expressed by the fuzzy sets HRR, RR, AHR, LR, HLR referrng to hard rght, rght, ahead, left, hard left. The output commands are normalzed n a scale from 0 to 1 for the translatonal velocty, where 0 corresponds to complete stop and 1 to maxmum speed. Rotatonal velocty output commands are normalzed from -1 to 1, where -1 corresponds to a rght turn wth maxmum angular velocty and 1 to a left turn wth maxmum angular velocty. 1502

5 System Inputs (Fuzzy Controller Inputs) (Fuzzy Control Outputs: Motor Commands) Control Reference Input (next Waypont) Current Target Waypont (lat, lon) Headng Error Calculaton Headng Angle Error (θe) Sensor Inputs GPS (lat, lon) Odometer (Robot Moblty Vrtual Encoder output) Laser scan Motorola Propretary GPS Fller Laser Preflter Fltered GPS Poston (lat, lon, angle) Robot Poston Detecton (uses current and past GPS readngs and odometry) Laser Range Flter (uses current and past laser scans and odometry) Fltered Laser Range Converson to Fuzzy nput varables: Left Range Center Range Rght Range Left Center Rght Controller Fuzzy Controller Rule Base Translatonal Velocty (v) Rotatonal Velocty (θ ) Robot Motor Drve System Plant (Feedback va Envronment) Fg. 3. Controller Block Dagram Each fuzzy rule j s expressed as: IF d 1 s D j1 AND d 2 s D j2 AND d 3 s D j3 AND he s HE j THEN tr s TR j AND rv s RV j ; for d for j=1,, number of rules. D j, s the fuzzy set n the jth rule whch takes the lngustc value of C,MD,A. HE j s the fuzzy set for the he whch takes the lngustc values FL, L, AH, R, FR. TR j and RV j are the fuzzy sets for tr and rv respectvely. The generc mathematcal expresson of the jth navgaton rule s gven by: µ ( j) ( d, hetr,, r) = mn[ µ ( d ), µ ( j) ( he), µ ( j) ( tr), ( j) ( rv)] µ R D HE TR RV The overall navgaton output s gven by the max-mn composton and n partcular : µ ( tr, vr) = max mn[ µ ( d, he), µ ( d, he, tr, rv)] N where d, he AND J R ( d, he, tr, rv) = µ ( ) R j= 1 µ ( d, he, tr, rv). The navgaton acton dctates change n robot speed and/or steerng correcton and t results from the deffuzfcaton formula, whch calculates the center of the area covered by the membershp functon computed from the last equaton. I. EXPERIMENTAL RESULTS Testng and valdaton of the prod MATLAB based fuzzy logc controller wthn the overall dstrbuted-sfx archtecture has been performed usng sngle and multple robots. Experments have been performed n an outdoor envronment (somewhat uneven terran) wth drt, grass, trees and some vegetaton. The frst set of R experments requred that the robots travel from an ntal poston to a dstant goal poston whle avodng many obstacles n unknown locatons (navgaton n an area wth trees). The second set of experments requred that robots follow sets of wayponts generatng search patterns. Patterns nclude raster scans wth two robots startng from dfferent ntal postons, avodng each other as well as other obstacles found n ther paths. All pattern and goal ponts are specfed n GPS coordnates A. Experment 1 The frst experment demonstrates ablty to move n an envronment wth many unknown obstacles (trees). The robot s gven an ntal poston on one sde of a large group of trees and a fnal goal pont locaton on the other sde of the group of trees. Fg. 4 shows one full path traveled through the tree covered area from the ntal pont to the fnal pont; perodc laser scans are shown over the course of the robot s path. Ths experment has been repeated several tmes, and Fg. 5 shows the dfferent paths followed by the robot for three repettons of the same experment. Lattude GPS Ponts Tree Locatons Robot Path Longtude Fg. 4. Robot path laser scans are shown Path dfferences are caused by varatons n the nput readngs produced by dfferences n GPS recepton, and 1503

6 by subtle and cumulatve dfferences n laser readng and odometer readngs. The outdoor envronment s effectvely contnuous; hence a small change n movement or ntal condtons may result n a dfferent path followed every tme. However, n all cases, the robot moves toward ts fnal goal GPS pont. Lattude Robot 1 Laser Scans Robot 1 Path Goal Pont Locaton Tree locatons Robot Path run 1 Robot Path run 2 Robot Path run 3 Robot 2 Path Robot 2 Laser Scans Lattude Fg. 7. Robots avodng each other Longtude W aypont Locatons Robot 1 Start Tree Locatons Robot 1 Path Robot 2 Start Robot 2 Path Robot 2 Fnsh Robot 1 Fnsh Fg. 5: Three dfferent paths for the same scenaro B. Experment 2 Experments have been performed wth two robots operatng smultaneously. Robots travel through a set of goal ponts delneatng a box shape. The robots start from dfferent ntal postons and travel to each of the goal ponts defnng the box. One robot moves clockwse, the other counterclockwse. The paths followed are shown n Fg. 6; smlar results have been obtaned repeatng the same experment. The robots negotate a statc object (a tree) and a dynamc object (the other robot) as they traverse the set of goal GPS ponts. Whle the robots were movng they crossed each other s path and avoded one another as shown n Fg. 7. Lattude Longtude W aypont Locatons Tree Locatons Robot 1 Path Robot 2 Path Longtude Fg. 6. Path followed by the two robots n the second experment C. Experment 3 A raster search s performed usng two robots. Robots start from dfferent postons n the feld and move n opposte drectons whle performng the raster search. The trajectores followed by the two robots are presented n Fg. 8. The experment has been repeated several tmes wth smlar results. Lattude Longtude Fg. 8. Raster search performed by the two robots II. CONCLUSIONS AND DISCUSSIONS Ths paper has presented a MATLAB based approach usng fuzzy logc controllers to robot navgaton n outdoor envronments usng fuzzy logc controllers. The mult sensor fuzzy logc robot controllers have been derved and mplemented usng laser, GPS and odometer data as nputs, also consderng fusng of such sensor data (laser, GPS and odometer) flterng out nose. Extensve outdoor envronment experments wth one and two moble robots have quantfed sensor naccuraces affectng navgaton and have demonstrated sngle and multple robot waypont navgaton, goal pont navgaton, random and raster scan search patterns along wth collson avodance. A major advantage of ths approach s that t may stand alone as a control archtecture, or t may be part of an overall mult layer dstrbuted system archtecture where data exchange, commands, recrutment, nterpretaton and processng of sensor fuson data s accomplshed at dfferent levels (obvously wth communcaton delays). The control system may contnue functonng under communcaton loss wthout affectng robot moton. Future research wll nclude controller enhancement nvolvng addtonal sensors lke IMU, IR and standard vdeo; dervaton of remote supervsory controllers for coordnated movement and varable robot autonomy; 1504

7 nter robot communcaton and mult robot dstrbuted control. REFERENCES 1. Saffott, A., Handlng uncertanty n control of autonomous robots, n Uncertanty n Informaton Systems, edted by A. Hunter and S. Parsons, Sprnger, LNAI 1455, pp: , Murphy, R. R., Introducton to AI Robotcs, MIT Press, Arkn, R. C., Behavor-Based Robotcs, MIT Press, Balch, T., Parker L. E., (Edtors), Autonomous Robots, A. K. Press, Dotsds, L., Valavans, K. P., Tsourvelouds, N. C., Fuzzy Logc Based Autonomous Skd Steerng Vehcle Navgaton, In Proceedngs of the ΙΕΕΕ Internatonal Conference on Robotcs and Automaton, pp: , Washngton DC, Valavans, K. P., Hebert, T., Kolluru, R., Tsourvelouds, N. C., Moble Robot Navgaton n 2-Dynamc Envronments Usng Electrostatc Potental Felds, IEEE Transactons on Systems, Man and Cybernetcs, Vol. 30, No. 2, Part A, pp: , Tsourvelouds, N. C., Valavans, K. P., Hebert, T., Autonomous Vehcle Navgaton Utlzng Electrostatc Potental Felds and Fuzzy Logc, IEEE Transactons on Robotcs and Automaton, Vol. 17, No. 4, pp: , Abdessemed, F., Benmahammed, K., Monacell, E., A fuzzybased reactve controller for a non-holonomc moble robot, In Robotcs and Autonomous Systems, artcle n press. 9. Lee, T. L., Wu, C. J., Fuzzy Moton Plannng of Moble Robots n Unknown Envronments, Journal of Intellgent Robotc Systems, Vol. 37, pp: , Agure, E., Gonzalez, A., Fuzzy Behavors for Moble Robot Navgaton: Desgn, Coordnaton and Fuson, Internatonal Journal of Approxmate reasonng, 25: , Goodrdge, S. C., Luo, R. C., Fuzzy Behavor Fuson for Reactve Control of an Autonomous Moble Robot: MARGE, In Proceedngs of the IEEE Internatonal Conference on Robotcs and Automaton, pp: , San Dego, CA, Ishkawa, S., A Method of Indoor Moble Robot Navgaton by Usng Fuzzy Control, In Proceedng of the IEEE/RSJ Internatonal Workshop on Intellgent Robots and Systems IROS, pp: , Osaka, Japan, L, W., Fuzzy Logc-based Percepton-Acton Behavor Control of a Moble Robot n Uncertan Envronment, In Proceedngs of the 3 rd IEEE Conference on Fuzzy Systems, Vol. 3, pp: , Pn, F. G., Watanabe, Y., Navgaton of Moble Robots Usng Fuzzy Logc Behavorst Approach and Custom Desgn Fuzzy Inference Boards, Robotca, Vol. 12, Saffot, A., Fuzzy Logc n Autonomous Robotcs: Behavor Coordnaton, In Proceedngs of the IEEE Internatonal Conference on Fuzzy Systems, pp: , Saffot, A., Konolge, K., Ruspn, H. E., A Multvarable Logc Approach to Integratng and Plannng and Control, Artfcal Intellgence, 76: , Tunstel, E., Moble Robots Autonomy Va Herarchcal Fuzzy Behavor, In Proceedngs of the 6 th Internatonal Symposum on Robotcs and Manufacturng, pp: , Seraj, H., Howard, A., Behavor-Based Robot Navgaton on Challengng Terran: A Fuzzy Logc Approach, IEEE Transactons on Robotcs and Automaton, Vol. 18, No. 3, pp: , Hagras, H., Callaghan, V., Colley, M., Outdoor Moble Robot Learnng and Adaptaton, IEEE Robotcs and Automaton Magazne, pp: 53-69, Mal, A. D., On the Behavor-Based Archtectures of Autonomous Agency, IEEE Transactons on Systems, Man and Cybernetcs, Part C: Applcatons and Revews, Vol. 32, No. 3, pp: , August Borensten, J. and Feng. L., Measurement and Correcton of Systematc Odometry Errors n Moble Robots, IEEE Journal of Robotcs and Automaton, Vol. 12, No. 6,, pp: , December Ojeda, L. and Borensten, J., 2003, "Reducton of Odometry Errors n Over-constraned Moble Robots," In Proceedngs of the UGV Technology Conference at the 2003 SPIE AeroSense Symposum, Orlando, FL, Aprl 21-25, Panzer, S., Pascucc, F., Ulv, G., An Outdoor Navgaton System Usng GPS and Inertal Platform, IEEE Transactons on Mechatroncs, Vol. 7, No. 2, pp: , June Ohno, K., Tsubouch, T., Shgematsu, B., Outdoor Navgaton of a Moble Robot between Buldngs based on DGPS and Odometry Data Fuson, In Proceedngs of Internatonal Conference n Robotcs and Automaton, pp: , Tape, Thrapp, R., Westbrook, C., Subramanan, D., Robust Localzaton algorthms for an autonomous campus tour gude, In Proceedngs of the IEEE Internatonal Conference on Robotcs and Automaton, Vol. 2, pp: , Vaneck, T.W., "Fuzzy Gudance Controller for an Autonomous Boat," IEEE Control Systems Magazne, pp: 43-51, Aprl, Bruch, M.H., Glbreath, G.A., Muelhauser, J.W.,and J.Q. Lum, "Accurate Waypont Navgaton Usng Non-dfferental GPS," AUVSI Unmanned Systems 2002, Lake Buena Vsta, FL, July 9-11, Long, M. T., Murphy, R. R., Parker, L. E., Dstrbuted multagent dagnoss and recovery from sensor falures, n Proceedngs of the 2003 IEEE/RSJ Internatonal Conference on Intellgent Robots and Systems, Vol. 3, pp: , Nelson, A. L., Dotsds, L., Long, M. T., Valavans, K. P., Murphy, R. R., Incorporaton of Matlab nto a Dstrbuted Behavoral Robotcs Archtecture, In proc. of the 2004 IROS/RSJ conference, pp: , S. Müller, H. Waller, Effcent Integraton of Real-Tme Hardware and Web Based Servces Into MATLAB, ESS'99 11th European Smulaton Symposum and Exhbton, Erlangen- Nuremberg, 1999, ESS'99, Oct , JMatLnk download ste: L. E. Parker, ALLIANCE: An Archtecture for Fault Tolerant Mult-Robot Cooperaton, IEEE Transactons on Robotcs and Automaton, Vol. 14, No. 2, pp: , A.L. Nelson, E. Grant, T.C. Henderson, Evoluton of neural controllers for compettve game playng wth teams of moble robots, Journal of Robotcs and Autonomous Systems, Vol. 46, No. 3, pp: , Mar G. J. Barlow, T. C. Henderson, A L. Nelson, E. Grant, Dynamc Leadershp Protocol for S-nets, Proceedngs, 2004 IEEE Internatonal Conference on Robotcs and Automaton (ICRA), New Orleans, LA, Aprl R. R. Murphy, R. C. Arkn, Sfx: An Archtecture For Actonorented Sensor Fuson, Proceedngs of IEEE/RSJ Intellgent Robots and Systems, Vol. 2, July 7-10, pp: , N. N. Okello, D. Tang, D. W. McMchael, TRACKER: a sensor fuson smulator for generalzed trackng, Proceedngs of Informaton, Decson and Control, IDC, pp: , February R. Wllgoss, V. Rosenfeld, J. Bllngsley, Hgh precson GPS gudance of moble robots, Proceedngs of the Australasan Conference n Robotcs and Automaton, M. T. Long, Creatng A Dstrbuted Feld Robot Archtecture for Multple Robots, M.Sc. Thess, USF, Summer J. Contreras, A. Los, M. Russo, JAVA/MATLAB smulator for power exchange markets, Power Industry Computer Applcatons, PICA Innovatve Computng for Power - Electrc Energy Meets the Market. 22nd Internatonal Conference on IEEE Power Engneerng Socety, pp: May 20-24,