URBAN NAVIGATION SYSTEM FOR AUTOMOTIVE APPLICATIONS USING HSGPS, INERTIAL AND WHEEL SPEED SENSORS

Transcription

1 URBAN NAVIGATION SYSTEM FOR AUTOMOTIVE APPLICATIONS USING HSGPS, INERTIAL AND WHEEL SPEED SENSORS Marana Spangenberg, TéSA/Thales Alena Space Olver Julen, Unversty of Toulouse, ENAC Vncent Calmettes, Unversty of Toulouse,ISAE-SUPAERO/TéSA Grégore Duchâteau, Thales Alena Space BIOGRAPHY Marana Spangenberg s a PhD student snce 27 and her work s jontly support by TéSA laboratory and Thales Alena Space, n Toulouse, France. She has graduated as an Electronc Engneer from the Unversty of Buenos Ares (Argentna). Her man areas of nterest are the GPS urban navgaton, usng dead reckonng sensors for obtanng hgh precson solutons. Vncent Calmettes receved Ph.D. degree n sgnal processng from SUPAERO (ENSAE), Toulouse, France. He s the head of the Laboratory of Electroncs and Physcs, at SUPAERO. Hs research nterests nclude the development of solutons based on DSP and programmable logc devces or ASICs for applcatons n Dgtal communcatons and Sgnal processng. He s currently workng on new Galleo sgnal processng and GNSS recevers archtecture. He s also nvolved n several projects devoted to postonng and atttude determnaton, ncludng low cost MEMS sensors characterzaton and INS/GPS ntegraton. Olver Julen s an assstant professor at the sgnal processng laboratory of ENAC (Ecole Natonale de l Avaton Cvle), Toulouse, France. Hs research nterests are GNSS recever desgn, GNSS multpath and nterference mtgaton and GNSS nteroperablty. He receved hs B.Eng n 21 n dgtal communcatons from ENAC and hs PhD n 25 from the Department of Geomatcs Engneerng of the Unversty of Calgary, Canada. Grégore Duchâteau s graduated from the Aeronautc and Space Engneerng school Supaero n Toulouse (France). He worked for Alcatel that he joned n 1991 and where he has worked n several areas related to fault tolerance, supervson and control for both wreless telecommuncaton and transport (ral, road, space) applcatons. Snce 23 he s techncal manager of the Transport soluton developed by the navgaton busness unt of Thales Alena Space France and s n charge of the techncal roadmap of the Multservce Trackng Platform product. ABSTRACT Land vehcle navgaton n urban areas, where maskng effects are very frequent, s a major challenge for both the accuracy and the ntegrty of GPS-only soluton. Several strong effects lnked to urban canyon envronments can serously degrade the fnal poston soluton. Thus, the sole use of a GPS to navgate n urban areas has been proven to be challengng when hgh performances are expected n terms of accuracy and ntegrty of the computed soluton. Ths paper proposes a Smooth correcton functon appled to the GPS receved sgnals. Ths functon works on the relablty of the pseudo-range measurements when affected by dverse sources of errors. Thus, to deal wth dfferent phenomena such as multpath or low C/No sgnals, a mergng technque between the pseudo-range and Doppler measurements s proposed. The couplng of GPS wth other postonng systems has been shown to be part of a soluton to overcome satelltes vsblty problem n urban areas. The dea s to combne external measurements together wth GPS measurements n order to mprove the overall navgaton system performance even durng complete GPS sgnal outages. Dead-reckonng sensors are a classcal soluton to ths problem. Nevertheless, due to the lack of stablty of these sensors, GPS s stll requred to correct ther correlated errors and gve an absolute poston when avalable. Classcal land vehcle navgaton solutons perform GPS augmentaton by usng Wheel Speed Sensors (WSS) and

2 gyros. Ths paper looks at the potental benefts of a couplng approach between these two types of sensors for a complete land vehcle navgaton soluton. WSS provde wheel angular rate measurements and are fundamental components n the antlock brakng system (ABS). Fnally, ths paper ntroduces a mult-aded navgaton system specfcally desgned for land vehcle navgaton n urban envronments. Not only sensors partcularly developed for land vehcle navgaton are tested, but also an nterestng use of the receved GPS sgnals (pseudorange and Doppler) s done, leadng to an optmzed tght ntegraton scheme. Test campagns are conducted to valdate the performance of the proposed ntegrated land vehcle navgaton system. The contrbuton of the system s evaluated for dfferent scenaros (urban and open areas). INTRODUCTION The new low-cost locaton technologes have rendered navgaton systems affordable for many cvl users. Among the fast growng GPS market, n-car navgaton s becomng a leadng applcaton ether for personal or professonal use. However, a wde horzon of new servces combnng postonng systems and other already exstng servces s beng explored. Several land vehcle postonng applcatons are requrng hgh precson solutons. The transportaton of dangerous goods or road tollng are matchng examples of navgaton soluton that requres both accuracy and ntegrty of the computed soluton. Navgaton n urban areas appears as one of the most challengng scenaros for performng poston estmatons usng just the GPS. In fact, the receved sgnal wll be serously affected by dfferent urban canyon phenomena that degrade the qualty of the receved sgnal: The lne-of-sght (LOS) of several satelltes can be obstructed, leadng to potental strong reductons of the receved C/N and regularly to the loss of several measurements. In the latter case, ths can lead to the mpossblty to compute a poston or to a strong value of the DOP. The ncreasng proporton of multpath that, due to strong reflecton, can have an assocated power close or even stronger than the drect LOS. Ths causes mportant rangng errors that wll affect the poston soluton. The potental hgh dfference of receved power between obstructed and non-obstructed satelltes can lead to cross-correlaton problems resultng n erroneous measurements. In the followng subsectons the last two effects wll be referred as measurement errors. Ths generalzaton wll be done because n real lfe t s very dffcult to dentfy and separate each of these phenomena (a decrease on the qualty of the sgnal s observed, but ts specfc cause can dffcultly be known). The nstantaneous presence of these dfferent sources of errors s very hard to predct and quantfy, and therefore ther correcton s one of the man urban navgaton problems. Many studes have been conducted proposng dfferent approaches to address ths problem. The frst approach proposed n ths paper s the development of a HSGPS-based recever algorthm. Havng GPS sgnals avalable n normal and very nosy condtons, a poston soluton wll be provded. The basc concept wll rely on the fact that the Doppler, though t provdes a correcton over the veloctes, should be less affected by the measurement errors than the absolute pseudo-range measurement [1]. Therefore, when the confdence over ths latter measurement doesn t reach a certan threshold, t wll be consdered to be corrupted by measurement errors. Then, a Doppler-smoothed pseudo-range or just the Doppler measurement wll be used to correct the state vector parameters. The decson parameter to establsh the degree of confdence over the receved pseudo-range measurement wll be based on RAIM (Recever Autonomous Integrty montorng) algorthm prncples [5]. The detecton stage of the ntegrty procedure s smlar to the one performed by the hgh precson OEM4 Inertal soluton developed by Novatel [2]. However, the man dfference/enhancement reles on the excluson level. Once a corrupted pseudo-range measurement s detected, the correspondng GPS satellte won t be dscarded, but t wll be exploted by usng a dfferent type of GPS measurement as stated on the prevous paragraph. Ths approach s mostly sutable for urban navgaton where measurements redundancy s scarce. On the other hand, when the contnuty over the trackng soluton must be guaranteed, a well-known approach s to put on board dead reckonng sensors that wll help the navgaton flter to manly cope wth GPS outages. Dead reckonng technques compute the actual poston of the vehcle from ts known ntal locaton by ntegratng dfferental measurements. Among the most used dead reckonng systems, we fnd the Inertal Navgaton System (INS). INS [3] are self contaned non jammable systems consstng of 3D accelerometers and gyros. At the begnnng ths soluton was restrcted to the aeronautcs market due to ts hgh cost. Nowadays, wth the development of low-cost sensors, such as MEMS, ths soluton s becomng wdely spread. Nevertheless, t must be ponted out, that these sensors have low precson, so a

3 low-cost stand-alone INS navgaton soluton wll have very poor performances durng GPS outages. In order to face the typcal aforementoned ssues, and focusng on land vehcle navgaton, ths paper nvestgates the use of the WSS (Wheel Speed Sensor). WSS are fundamental components n the antlock brakng system (ABS), whch s standard equpment n all new generaton vehcles. WSS provde wheel angular rate measurements through the vehcle Controller Area Network (CAN). The rotaton rates from dfferent wheels can be processed to get the headng, the velocty and then the dstance travelled by the vehcle. As almost all modern cars are now equpped wth such sensors, ther use as an alternatve source of navgaton may be very effcent at no addtonal cost. However small errors n the WSSs measurements can ental sgnfcant bases n the headng s angle calculaton. A coupled WSS/gyro schema s consdered an nterestng navgaton soluton enablng the trackng of the vehcle durng longer perods. The navgaton system desgned and tested n the paper s composed of three parts: a Hgh Senstvty GPS recever, WSSs and a gyro. All these sub-systems are ntegrated accordng to a tght couplng archtecture n order to take beneft of any avalable measurement. The ntegrated navgaton system s buld around an adaptve central Extended Kalman Flter (EKF) capable of estmatng the dead-reckonng sensors errors usng raw GPS measurements (.e. pseudoranges and Dopplers). Ths paper s organzed as follows: Secton I presents HSGPS-based flterng algorthm usng a smooth correcton functon that merges pseudo-range and Doppler measurements. WSSs and gyros functonng prncples are explaned n Secton II. Secton III presents the flterng couplng approaches for the navgaton systems of the two prevous sectons. Expermental results are shown on Secton IV. Fnally, conclusons are drawn. I GPS RECEIVER ALGORITHM TYPES OF GPS MEASUREMENTS The three GPS measurements obtaned from the recever are: Pseudo-range Carrer Phase Carrer Doppler The pseudo-range measurements provde absolute postonng but are strongly affected by multpath. On the contrary, the phase measurements provde relatve postonng (the ambguty s needed n order to calculate the present poston) but have the most mportant advantage of beng hghly accurate. However, f a certan level of contnuty over the GPS recepton s not guaranteed, the phase measurement wll be affected by cycle slps. Ths paper addresses the navgaton problems present n urban scenaros where sgnals are hghly nterfered and GPS avalablty s scarce. Therefore, the carrer phase wll not be consdered a sutable measurement when navgatng n these scenaros. The thrd type of GPS measurement dscussed n ths paper s the Doppler measurement. Knowng the satelltes poston and velocty from ephemers messages, ths measurement enables the estmaton of the vehcle s veloctes, and hence by ntegraton, t provdes a relatve poston. At the same tme, ts most nterestng advantage reles on ts theoretcal robustness to degraded scenaros n partcular to multpath. For ths approach to be vald, t s consdered that Doppler measurements are not derved from the pseudo-ranges. Ths assumpton was valdated for at least one type of recever accordng to [1]. Consderng the aforementoned characterstcs for the three types of GPS measurements, the followng approach wll be dscussed n ths paper: degraded pseudo-range measurements, mostly affected by urban canyon phenomena, may be complemented by more robust Doppler measurements. In partcular, f goodvsblty/low-nterference GPS condtons are present, the pseudo-range measurement can be used for the poston soluton. Otherwse, f one of these two condtons s not satsfed, the Doppler measurement may be ntroduced n the measurement model. PSEUDO-RANGE & DOPPLER MODELS The EKF s used as the navgaton flter. In eq. (1) and (2), the flter state space model s recalled: X( t) = F( t) X( t 1) + v( t) y( t) H( t) X( t) + u( t) where X(t) s the state vector, F(t) s the state model propagaton matrx, y(t) s the measurement vector and H(t) s the EKF lnearzed non lnear functon relatng the measurements to the state vector. The nose sequences v(t) and u(t) are supposed to be ndependent whte Gaussan wth covarances matrces Q(t) and R(t). As concluded n the prevous secton, the pseudo-range and Doppler measurements are gong to be used for the navgaton soluton. Ther models are presented n equaton s (3) and (4). (1) (2) [ ] [ ] [ ] PR y ( t) = X ( t) x( t) + Y ( t) y( t) + Z ( t) z( t) + b( t) + w ( t) (3)

4 L1 y ( t) = - (( V ( t)- V ( t)) + ( V ( t)- V ( t)) + ( V ( t)- V ( t))) LOS c L1 (4) - d ( t ) + n ( t ) c DOP X X Y Y Z Z sat / veh for =1,...,ns, where ns s the number of vsble satelltes, y PR (t) s the pseudo-range between the user and the th satellte, [X(t),Y(t),Z(t)] T s the poston of the th satellte n the local frame, [x(t), y(t), z(t)] s the vehcle poston to be estmated, b(t) s a bas term resultng from the clock offset, and w (t) are the measurement errors such as recever nose, onospherc, tropospherc and ephemers errors. On the other hand, the y DOP s the Doppler shft cause by the relatve moton of the vehcle wth respect to the th satellte, projected nto the unt lne of sght vector (LOS sat/veh ) between the satellte and the vehcle. The th satellte veloctes are denoted by [V x (t), V y (t), V z (t)], the users veloctes by [V x (t), V y (t), V z (t)], d(t) s the drft n the clock offset, n (t) are the measurement errors, L1 s the carrer frequency equal to MHz and c s the speed of lght. These equatons wll be used as measurement equatons for estmatng the desred user poston. SMOOTH CORRECTION FUNCTION The man challenge when performng urban navgaton s to reduce the postonng bas ntroduced by the measurement errors. A well-known approach to deal wth these problems, n partcular wth multpath and measurement nose, s the Hatch flter [4]. The prncple of ths flter s the followng: the pseudo-ranges are merged wth the carrer phase measurements n a recursve flter that progressvely ncreases the weght of the carrer phase whle decreasng the weght of the pseudo-range measurement. The goal of ths method s to smooth the pure pseudo-range by the carrer phase measurement and eventually provde a smoothed range measurement that t s mostly obtaned from the accurate carrer phase measurement. The tme of transton between a hghly domnated pseudo-range and a hghly domnated phase measurement s gven by the flter smoothng tme. Ths convergence tme s set to be superor to the correlaton tme of multpaths. The Hatch flter, avalable n almost all recevers, can be appled f open-sky condtons nsure contnuty over the recepton of the GPS sgnals (t was manly conceved for open areas). Otherwse, as explaned n the prevous secton, the carrer phase measurement wll be affected by cycle-slps and early loss of lock n challengng envronment. Hence, ths flter cannot be consdered a sutable soluton for urban envronments. A navgaton technque takng nto account the approach of the Hatch flter but usng a dfferent smooth correcton crteron s proposed n ths paper. The robustness of the Doppler measurement s gong to be exploted. A mergng technque between the absolute pseudo-range and the Doppler measurement s proposed. The dea of assgnng weghts to the dfferent GPS measurements wll be addressed. In partcular, the weghts assocated to the pseudo-range and Doppler measurements are gong to be constantly calculated, and wll have almost no correlaton wth prevous samplng tmes. The mergng expresson for ths approach s presented n eq. (5). smooth PR c DOP y ( t) = wpr y ( ) ( 1) ( ) t + wdop y t y t dt L1 where w pr and w dop are the respectve pseudo-range and Doppler weghts, dt s the GPS samplng tme, sub-ndex denotes the th satellte n lne of sght, and y (t-1) s the prevous absolute measurement. The correspondng nose of the smoothed measurement s also a lnear combnaton between the assocated noses of the pseudorange and Doppler measurements. The weghts wll be chosen accordng to the values obtaned from the calculaton of the EKF nnovaton (6) and nnovaton covarance (7). PR PR Innov = y ( t) yˆ ( t) T Innov _ cov = H( t) P H( t) + R( t) t / t 1 PR where yˆ ( t) s the estmated pseudo-range measurement from the state vector propagaton, and P t / t-1 predcted estmate covarance. The values obtaned from equaton (6) and (7) wll be taken as ndcators of the receved pseudo-range qualty and therefore, of ts capablty to correctly estmate the poston. It must be ponted out that we are gong to consder as the man source of error for the flter model the ncorrect calculaton of R(t): nether the model assocated to the state transton matrx nor to the measurements transton matrx ntroduce sgnfcant errors. The dea s that the pseudo-range covarance nose matrx R(t) wll be constructed from the measurements noses n non corrupted scenaros. If the characterstcs of the recever and the C/No of the receved sgnal are known, these values can be correctly estmated for typcal envronments. Assumng no temporal correlaton n the pseudo-range measurements and no correlaton among the recever channels, the R(t) matrx wll be a dagonal matrx contanng the nose varance for each sgnal. However f the sgnal s affected by any knd of measurement errors, as these are not detected a pror by the recever and therefore cannot be quantzed, the R(t) matrx wll not take them nto account. As a consequence, the proposed pseudo-range measurement model wll be no (5) (6) (7)

5 longer vald: the assocated sgnal nose varance wll be smaller than ts real value. The prncple to decde whether or not the pseudorange measurement s relable accordng to the values gven by eq. (6) and (7), s based on a least-square resdual type method commonly used for RAIM (Recever Autonomous Integrty Montorng) technques [5]. Ths algorthm uses an overdetermned system to perform the consstency check, where 5 satelltes are needed at least to detect a faulty measurement and 6 to be able to exclude t. In urban areas, such a hgh vsblty over the satelltes constellaton s very rare. Hence a RAIM approach s not a sutable soluton. Ths paper proposes an adapted and smplfed urban ntegrty check. In our case, the module of each nnovaton obtaned from the EKF wll be compared to ts assocated probablty dstrbuton functon and the detecton and solaton wll be performed smultaneously. The nnovatons dstrbuton functon wll be characterzed by a central gaussan dstrbuton. As for the RAIM detecton stage, the decson threshold wll depend on the false detecton probablty (FDP). Usng an adequate threshold crteron [6], the FDP s set to 6.33 e- 5. Then the decson threshold for each th measurement wll be gven by: = 4. λ σ (8) FDP nnov Where σ nnov_ s the standard devaton assocated the th fltered nnovaton. If ths threshold s exceeded, the pseudo-range measurement wll be excluded from the flterng model and wll be replaced by ts correspondng Doppler measurement. A second lower threshold wll be statstcally establshed to perform the so called Smooth correcton usng (5). Ths threshold s set to: λ = 2. σ (9) Smooth nnov Therefore, wthn a 2. σ nnov_ devaton the pseudo-range measurement wll be consdered totally useful and not corrupted. Ths accounts for an equvalent FDP of.4 (t can t be formally consdered a FDP because the pseudorange measurement won t be excluded beyond ths threshold but rather merged wth the Doppler measurement (5) ). Probablty densty functon Test statstcs for Innovaton's module 2. σ nnov PR correcton 4. σ nnov Smooth correcton Doppler correcton λ Smooth λ FDP Fgure 1 Test Statstc for EKF nnovaton (5) A smooth correcton functon (fgure 1) assocated to the relablty on the receved pseudo-range measurement s proposed. If the pseudo-range flter model (assocated to the calculaton of R(t) ) s proven not to be vald, the Doppler measurement s ntroduced as a complementary/alternatve source of measurement. PR weght (wpr) PR correcton σ 4. nnov σ nnov Innovaton absolute value Fgure 2 PR weghts assocated to the Smooth correcton crteron Fgure 2 llustrates the functon whch wll determne the weght values (mportance or relablty values) assocated to the each pseudo-range measurement, dependng on the actual rato between eq. (6) and (7). A weght value equal to 1 ndcates that the pure pseudorange s used n the flter measurement model, whle a value equal to means that no contrbuton of the pseudorange s consdered. The relatonshp between the pseudorange and Doppler weghts s: w = 1 w (1) DOP Smooth correcton where the pseudo-range weght decreases lnearly between 2 and 4 σ nnov. The crteron for decdng whch type of GPS measurement s gong to be use as an nput to the navgaton flter s dvded n three categores: If the nnovaton (eq. (6)) over the th recever channel has a value smaller or equal to 2 tmes ts correspondng standard devaton σ nnov (obtaned from the th element of the dagonal of eq. (7)), the flterng model s consdered correct, and the absolute pseudo-range measurement s used to correct the poston. If the nnovaton over the th recever channel has a value that s bgger or equal to 4 tmes ts correspondng standard devaton σ, the nnov pseudo-range measurement s consdered hghly PR Doppler correcton

6 corrupted, and the more robust Doppler measurement s used to correct the veloctes. If the nnovaton over the th recever channel has a value that s between 2 and 4 tmes ts correspondng standard devaton σ, a nnov smoothed measurement s used. Ths measurement s obtaned from the pseudo-range smoothed by the Doppler measurement accordng to ther respectve assocated weghts (eq. (5)). Some partcular cases for the applcaton of ths smooth correcton crteron are consdered n the followng paragraph. If n the prevous nstant (t-1) the measurement used for the th recever channel was the Doppler measurement, then there s no th absolute measurement avalable for (t-1). Hence, the smooth correcton expresson of eq. (5) s not a possble opton at tme nstant (t): ether the pseudo-range (w pr =1) or the Doppler measurement (w pr =) s used for the flterng. At the same tme, for an average urban vehcle dynamcs (~ 3 km/h), the multpath correlaton tme s consdered smaller than 5 seconds, so consecutve correctons for the same th recever channel usng just the Doppler measurements can be performed for a perod not exceedng the 5 seconds. Indeed, f just the Doppler s used to correct the poston, a dvergent soluton over tme s obtaned; therefore the pseudo-range measurement s perodcally needed. II - DEAD RECKONING SENSORS Durng partal and total GPS outages the contrbuton of the dead reckonng (DR) sensors s crucal. They enable contnuty over the poston estmaton, though, f left alone for a long perod, they wll provde hghly based solutons. However, for short GPS outages perods, they can be used as a relable source of navgaton. Moreover, as they gve a pror nformaton of the vehcle s dynamcs, ntegrty checks as the one propose on the prevous secton, may become more effcent. Corrupted pseudo-range measurement are not compared to a statstcal model of the vehcles dynamcs, but to the soluton computed from the real DR measurements. In the followng subsectons, two dfferent types of automotve dead-reckonng sensors are presented. WSS (WHEEL SPEED SENSOR) Ths secton descrbes the man elements of dfferental odometry. The dea s that the dstance traveled and the yaw rate can be calculated usng the measurements gven by the vehcle WSSs. The WSS measurements wll be calbrated to provde relable nformaton durng GPS satellte outages. Fgure 2 shows WSSs located on the front and rear wheels. The frst ndex of the dfferent varables refers to the front f or rear r axes whereas the second ndex corresponds to the left l and rght r sdes of the car. Consequently, the wheel rad (resp. angular veloctes) are denoted as R rl,r rr, R fl and R fr (resp. w rl, w rr, w fl and w fr ). The other notatons used n fgure 2 are L for the length between wheels and for the vehcle yaw rate (change of angle of drecton). Ths paper focuses on velocty and yaw rate calculatons usng rear wheels. Ths choce s motvated by the fact that acceleraton and deceleraton have less effect on the output of these sensors.. Fgure 3 - Vehcle's model Assume frst that the wheel rad are constant and known. The mean speed of the vehcle (n the along track drecton) at tme t can be computed as [5]: v along _ track ωrr ( t) Rrr + ωrl ( t) Rrl ( t ) =. 2 (11) The yaw rate of the vehcle can be calculated smlarly as a functon of the angular veloctes of each wheel (equaton 11). By neglectng sde slp effects (see [7] for more detals) and modelng the vehcle as a rgd body, the vehcle yaw rate at tme t expresses as: ωrr ( t) Rrr ωrl ( t) Rrl ψɺ ( t) =. L (12) Therefore, a two-dmenson poston soluton can be computed usng (13) (14), where the ne sub-ndex denotes the north-east frame. The fnal geodetc poston (lattude and longtude) wll be gven by λ( t) = λ( t 1) + ɺ λ( t). dt and φ( t) = φ( t 1) + ɺ φ. dt, where dt denotes the samplng perod. v along_track ( t).cos( ψ( t)) v ne( t) = v along_track ( t).sn( ψ( t)) (13)

7 (14) Errors n the wheel rad or n the measurements gven drectly by the sensor wll have a strong mpact on localzaton accuracy when gettng propagated through the angle and velocty expresson. Indeed, any non corrected error wll result n an accumulatve ncreasng error due to the ntegraton performed to obtan the poston. By couplng the WSSs wth the GPS, these errors can be reduced. Moreover, n secton IV, t s shown that the precson of the yaw angle calculated from the WSS measurements s very poor. A gyroscope wll be ntroduced to the dead-reckonng scheme, n order to enhance the navgaton performance. GYROSCOPE 1 ɺ λ = R ɺ φ λ v 1 Rφ cos( λ) ( t), Gyroscopes are nertal sensors that measure the rotaton rate of the sensor cluster. By ntegraton, the atttude angles can be calculated and the vehcle s orentaton can be tracked. Usng an addtonal sensor that delvers the dstance traveled, such as the WSS, the trajectory can be computed. In ths paper, a one-axe gyroscope wll be used to measure the headng or yaw angle. III - HYBRID NAVIGATION SYSTEMS HSGPS/WSS The common GPS/Dead-Reackonng (DR) couplng technques use the GPS pseudo-range measurements to correct the DR errors n order to obtan relable poston estmaton. The errors of the DR system are defned as the dfference between the actual DR parameter values and ther estmatons δx = X X estmated. The state model descrbng the WSS errors dynamc behavor can be obtaned by lnearzng the deal equatons around the WSS estmates. The state vector s then augmented wth systematc sensor errors: T 8 ( δλ, δφ, δ, δ, δψ, δψɺ,, ) R x = v v b d r l (15) where δλ and δφ contan the poston (lattude and longtude) errors, δv r and δv r are the velocty errors correspondng to the rght and left wheel, δψ and δψɺ are the yaw angle and yaw rate errors, b denotes the GPS recever clock offset n meters and d s ts drft. The terms δv r and δv l represent both the WSS errors and errors n the nomnal value of the wheel radus. Ths type of representaton was chosen because these two types of errors can be consdered addtve and ndependent for each wheel. Moreover, t cannot be granted that, f treated ne separately, these two sources of errors are observable. The measurement equatons for the HSGPS/WSS system are gven by eq. (3), (4) and (5). A frame transformaton must be done from the DR corrected geodetc coordnates (λ, Φ) to the rectangular coordnate system (x, y, z). For our two-dmenson navgaton system, the heght wll be consdered known. Ths knd of hybrd approach can be consdered as the most cost-effectve scheme for automotve navgaton augmented systems. However, due to the dfferental nature of the headng angle value (dfference between the two wheels velocty), f any slppng or skddng take place, the headng angle wll be affected by a very large error. A popular soluton to ths problem s to augment the system wth a gyroscope. HSGPS/WSS/Gyro Accordng to [6] and to the results shown n the next secton, the calculaton of the yaw rate from the WSS measurement shows serous lmtatons. Therefore, for the HSGPS/WSS/Gyro approach, the gyro wll provde the yaw-rate and the WSSs wll just be used to compute the along-track velocty. The flterng schema s very smlar to the one used for the HGPS/WSS system. The state augmented vector s formed by the followng parameters: T 7 ( δλ, δφ, δ, δψ, δψɺ,, ) R x = v b d (16) Where the only dfference wth (15) s that δψ and δψɺ correspond to the errors ssued from the gyro, and the WSSs errors are condensed nto δv (there s no need to dfferentate them because to the along-track velocty they all appear as one general, addtve error). For ths ntegraton approach no correlaton s consdered between the yaw angle error and velocty error. Algnment errors between the gyro and the vehcle s man axs are consdered neglgble. Fnally, the ntegrated navgaton soluton s presented n fgure 3. Three man blocs that act as an nput to the EKF may be dstngushed: 2 correspondng to the dfferent dead reckonng sensors (WSS and gyro), and the thrd one correspondng to the HSGPS-based recever algorthm where the Smooth correcton crteron s mplemented. WSS Sensors errors to be estmated Gyro Sensors measurements EKF Smooth correcton crteron PR Doppler Estmated errors Measurements that correct the sensors errors Fgure 4 - Integrated navgaton flter Poston estmaton

8 IV - EXPERIMENTAL RESULTS Ths secton presents the results of the measurement campagns conducted n order to test and valdate the proposed navgaton solutons. Two knds of envronments are studed: open-sky and urban scenaros. Specal attenton s pad to the performance of the dfferent navgaton approaches when tryng to deal wth corrupted sgnals (.e. measurement error) or GPS outages. Frstly, the HSGPS-based approach wll be tested, and later the dead reckonng sensors wll be ncorporated to acheve the fnal navgaton soluton. The reference trajectores for both scenaros are shown n yellow n fgure 5 and fgure 6. Sgnfcant ntervals of these trajectores are chosen to clearly llustrate the characterstcs and performances of the dfferent navgaton flters. Intervals of 2 mnutes are consdered and, n partcular for the urban scenaro, an nterval presentng dense urban canyon phenomena s chosen. Fgure 5 Open-sky reference trajectory Fgure 7 - GPS Samplng tme Fgure 6 Urban reference trajectory The Ublox TIM-LR GPS recever was used. The pseudorange and Doppler measurements were processed at 4Hz. Doppler measurements are fltered by the recever so as not to account for nstantaneous errors that could ental a hghly dvergent soluton. The onospherc and tropospherc errors were corrected usng the Egnos messages n post-processng, smulatng a real-tme lnk obtaned through a communcaton devce. To calculate the measurement nose covarance matrx R(t), the Egnos standard devatons for the atmospherc correctons were added to the computed (C/No)-based measurement errors. Fgure 7 llustrates the performances of four dfferent navgaton flterng technques. The frst one s a typcal least-square (LS) approach presented here as a standard benchmark soluton. The dfference between the last three technques (based on the EKF) reles on the measurement models and crterons used. For the curve n blue, a standard navgaton soluton usng pseudorange (PR) measurements s tested. On the other hand, the PR- Doppler flterng and Smooth flterng rely on ntegrty controls and varous measurement models. For the PR- Doppler flterng a standard sngle-threshold ntegrty check wll be used. The value of ths threshold s set to 4.σ nnov. After computng the nnovaton from the pseudorange measurements, ts value s compared to the establshed threshold. If the measurement s determned to be unrelable, t s replaced by ts correspondng Doppler measurement. The Smooth flterng s based on the proposed Smooth correcton functon : the prncple of ths algorthm s llustrated on fgure 1 and 2. Two thresholds are set for ths last enhanced approach and three types of measurement are used. Durng the frst 1 seconds of the trajectory n fgure 7 all the flterng approaches have the same performance. The qualty of the receved pseudorange measurement s very good so no detecton s computed for the ntegrty check. Fgure 8 shows the dfferent types (4) (5) and number of measurements used by the last two technques at every nstant, f the ntegrty check detects a corrupted pseudorange sgnal. For the last 6 seconds, two new satelltes are n vsblty wth very low elevaton angles and therefore very susceptble to measurement errors. Ther contrbuton degrades the performance f no ntegrty montorng s done. The Smooth flterng acheves the best performance by a mnmum mean error and a better stablty. Poston error [m] LS flterng PR flterng PR-Doppler flterng Smooth flterng Fgure 7 - HSGPS based navgaton soluton for open-sky scenaros

9 1.5 # of Doppler measurements used PR-Doppler flterng Smooth flterng # of Smooth measurements used 1 Poston error [m] Measurement errors PR flterng PR-Doppler flterng Smooth flterng 2 Sat 3 Sat The same flterng technques are tested n an urban scenaro. Evdently, for any non optmal trackng condtons (.e. measurement error or GPS outages present) the LS approach can t provde a vald poston soluton (fgure 9). Fgure 1 presents just a zoomed vew over fgure 9 to enable the analyss of the EKF-based flterng technques. In ths case both ntegrty-based flters present the same results and outperformed the standard PR flterng. In the presence of mportant measurement errors the Doppler measurement wll be drectly used, so the behavor of both PR-Doppler and Smooth flterng technques s the same. Partal GPS outages (2 or 3 satelltes n lne of sght), appear durng a 4 seconds tme nterval, entalng sgnfcant poston errors. The soluton to ths lack of measurements redundancy s to ntegrate addtonal trackng sensors to the vehcle. The fnal enhanced navgaton soluton s augmented wth dead reckonng sensors as presented n Secton III. Poston error [m] Fgure 8 Alternatve GPS measurements used n open-sky scenaro LS flterng PR flterng PR-Doppler flterng Smooth flterng Measurement errors 2 Sat 3 Sat Fgure 9 HSGPS based navgaton soluton for urban scenaro Fgure 1 HSGPS EKF-based navgaton soluton for urban scenaro 1.5 # of Doppler measurements used PR-Doppler flterng Smooth flterng # of Smooth measurements used Fgure 11 Alternatve GPS measurements used n urban scenaro The used WSSs were those that are already on board and that serve manly to the ABS system. The data was recuperated through the on-board dagnostc (OBD) port usng an ELM327 devce (whch s an OBD to RS232 nterpreter). The velocty frames were produced and processed at 5Hz. For the gyro, two possbltes were explored: an already on board gyro present n most vehcles havng the ESP (Electronc Stablty Program) opton, and an external gyro specally developed for land vehcle navgaton (Melexs MLX969). Ths last gyro worked at 7Hz. No sgnfcant loss of nformaton was found when nterpolatng the Melexs-gyro data at 5Hz, so as to algn the frequency of both dead reckonng systems. The propagaton rate of the EKF was set to 5Hz followng the sensors frequency. Every tme GPS measurements become avalable, the propagated state vector (contanng the poston error) s corrected. Table

10 1 summarzes the nose values for each of the deadreckonng sensors: Gyro ESP Gyro Melexs WSS.2 /s HZ.2 /s Hz.8 m/s Table 1 Sensors nose values The stand alone dead-reckonng poston solutons are presented on fgure 12. Two dfferent ntegraton models were tested followng the prncples presented on Secton II. As expected for the WSS soluton, where both the velocty and headng angle are calculated from the WSSs, the ntegrated ncreasng error over tme strongly affected the headng angle. Hence, the nterest of couplng a gyro to the system was shown. WSS+External Gyro (.e. the MLX969 model) was the most performng soluton. Due to the poor performances obtaned for the on board ESP gyro, ths ntegrated soluton was dscarded for posteror treatment. WSS+Gyro approach contnues to provde a relable poston estmaton whle the ncreasng ntegraton error of the WSS can t be controlled. Whle the mplementaton of the Smooth flterng presents better results for the WSS hybrd system, no sgnfcant contrbuton s observed for the WSS+Gyro approach. Bascally, the WSS+Gyro schema presents a hgh precson n an already stand-alone confguraton (fg 11). Therefore, ts assocated nose value s very low compared to the one of the nosy GPS sgnals. Durng the flterng correcton stage, the EKF won t gve a sgnfcant weght to the GPS sgnals but t wll manly rely on the dead-reckonng (WSS+Gyro) soluton PR/WSS PR/WSS/Gyro Smooth/WSS Smooth/WSS/Gyro Poston error [m] sat 3 sat 1 Reference WSSs WSSs +ESP Gyro WSSs+External Gyro Fgure 12 - Dead-reckonng navgaton solutons Fnally, the complete enhanced hybrd approach was mplemented. Whle t was already shown that the Smooth correcton crteron was able to deal wth the so called measurement errors, the performance of the coupled dead-reckonng solutons regardng the GPS outages s especally studed. Two dead-reckonng approaches are used: the no-addtonal cost WSSs and the WSSs+Melexs gyro. Ther performances are compared usng the standard PR navgaton soluton and the proposed Smooth soluton. As the essental contrbuton of ths last optmzed navgaton scheme s focused on GPS outages problematc, just the performances for the urban scenaro wll be shown (fgure 13). Both hybrd systems present good performances durng a partal GPS outage of 3 satelltes. However, for 2 satelltes n vsblty, only the Fgure 13 - Fnal enhanced navgaton CONCLUSIONS Ths paper nvestgated dfferent approaches for land vehcle urban navgaton. Among the urban canyon problems two man categores were studed: the postonng errors caused by the measurement errors and those caused by GPS outages. To address the frst source of errors, HSGPS-based recever algorthm was proposed. A smooth correcton functon combnng the pseudo-range and Doppler measurements showed very nterestng results n nosy scenaros. On the other hand, to deal wth GPS outages, deadreckonng sensors were ncorporated. The ABS already on-board WSSs and an addtonal gyro were exploted. By usng the WSSs from left and rght wheels, the vehcle s dstance traveled and yaw rate could be estmated. A coupled dead-reckonng system was proposed. As a consequence the navgaton performance was hghly mproved. Expermental results showed that the fnal

11 Smooth//WSS/Gyro hybrd system s a promsng navgaton soluton for urban envronments. Further studes should be dedcated to formalze the lnear transton model between the pseudo-range and Doppler measurements. Improvements over the WSS model to nclude sde-slppng effect should be analyzed. A feld test campagn to generalze the behavor of dfferent vehcle s WSSs s wthn the perspectves the authors have for ths work. [9] J. Gao, M. G. Petovello, and M. E. Cannon, Development of precse GPS/INS/Wheel Speed Sensor/yaw rate sensor ntegrated vehcular postonng system, n Proc. of ION NTM-6, (Monterey, CA), Jan 26. ACKNOWLEDGMENTS Ths work was supported by Thales Alena Space. The authors would lke to thank Benot Prot from ISAE- SUPAERO, for hs techncal support durng the feld test campagns. REFERENCES [1] D. Kubrak, C. Macabau, M. Monnerat, Vehcular Navgaton usng a Tght Integraton of Aded-GPS and Low-Cost MEMS Sensors, Proceedngs of the ION NTM 26. [2] T. Ford, J. Neuman, M. Bobye, OEM4 Inertal: An Inertal/GPS Navgaton System on the OEM4 Recever, KIS 21,Banff, Alberta, Canada. [3] J. A. Farrell and M. Barth, The Global Postonng System and Inertal Navgaton. New York: McGraw Hll, 1999 [4] Hatch R., The Synergsm of GPS Code and Carrer Measurements, Proceedngs of thrd Internatonal Geodetc Symposum on Satellte Doppler Postonng, [5] B. W. Parknson, P. Axelrad, Autonomous GPS ntegrty montorng usng the pseudorange resdual, Global Postonng System, Red Book volume V, [6] J. Bullock, M. Foss, G. Geer, M. Kng, Integraton of GPS wth other sensors and Network assstance, Understandng GPS prncples and applcaton, Second edton, Chapter 9. [7] N. Svenzén, Real tme mplementaton of MAP aded postonng usng a Bayesan approach, Master s thess, Unversty of Lnköpng, Lnköpng, Sweden, 22. [8] M. Spangenberg, J. Y. Tourneret, V. Calmettes, Fuson of GPS, INS and Odometrc data for automatve navgaton, EUSIPCO 27, Poland.