Sojourner on Mars and Lessons Learned for Future Planetary Missions

981695 Sojourneron M arsand Lessons Learned forFuturePlanetary Rovers

Brian W ilcox and Tam Nguyen NASA's JetPropulsion Laboratory

ABSTRACT The sitelocations w ere designated by a hum an operator using engineering datacollected during previous On July 4, 1997, the M arsPathfinder spacecraft traversals and end-of-solstereo im ages captured by the successfullylanded on M arsinthe Ares Vallislanding lander IMP (Im ager for M arsPathfinder) cam eras. site and deployed an 11.5-kilogram m icrorover nam ed During the traversalsthe rover autonom ouslyavoided Sojourner.Thismicrorover accom plished itsprimary rock,drop-off,and slope hazards. Itchanged its course mission objectives inthe first 7 days, and continued to toavoidthese hazards and turned back tow ardits goals operatefora totalof83 sols(1 sol= M ars day = 1 Earth w henever the hazards w erenolonger inits w ay. The day + ~24 m ins)untilthe landerlostcom m unication w ith rover used "dead reckoning" counting w heel turns and Earth, probably due tolander batteryfailure. The using on-boardrate sensorsestimate position. Although microrover navigated to m any sites surrounding the the rover telem etryrecorded itsresponses to hum an lander, and conducted various science and technology driver com m ands in detail,the vehicle'sactualpositions experim entsusing itson-boardinstrum ents. were not know n untilexamination of the lander stereo im ages at the end of the sol.Acollection of stereo Inthis paper,the rover navigation perform ance is im ages containing rover tracks allow s reconstruction of analyzed on the basisofreceived rovertelem etry, rover the rover physicaltraversalpaththoughoutthe m ission. uplink com m ands and stereo im ages captured by the Since the primary purpose for a robotic vehicleon lander cam eras. Its physical traversalpathisredraw n anotherplanetistomovepreciselytotargets ofscientific from the stereo im ages containing tracks and is interest,the abilityofthe vehicleto sense and navigateto com pared w iththe rover-recorded path and the driver- precise locations isim portantto gauge. The accuracy of planned path. Implications fornext-generation planetary navigation of Sojourner and itsimplications for future roversare described, including the sub-1-Kg N anorover planetaryroversisthe subjectofthis paper. being builtby NASA to conduct asteroidexploration as partofthe Japanese M U SES-C sam plereturnmission THE ROVER and the large roverw iththe Athena payload w hich w illbe used as partoftheMars sam plereturnprogram . Sojourner(Figures 1 & 2)isasix-w heeled vehicle68cm long,48 cm w ide,and 28 cm high (with17cmground INTRODUCTION clearance). The body is builton the rocker-bogie chassis which,by use ofpassive pivotarms,allow s the vehicleto The Pathfinder spacecraftlanded on M arsonJuly4, maintainanalm ost constantw eightdistribution on each 1997,and the nextday the Sojournerroverrolled dow n a w heel on veryirregular terrain. As a result,Sojourner ram p ontothe surface and began itsexploration of the w as abletotraverse obstacles about1.5timesasbigas Mars environm ent near the lander.Themission called the w heels, since the rear w heelsare abletomaintain for the rover tomovetosites ofinterestnearlyeverysol traction even w hile pushing the frontw heelsintovertical to conduct science and engineering experim ents. steps hard enough to getlifting traction. This consistsof Equipped w ith navigation and articulation sensors and linkages, sixmotorized w heels, and four m otorized- vision cam eras,the rover carried outits dailytraversals steering m echanism s. The vehicle'smaxim um speed is autonom ously based on setsofdriver com m ands sent about 0.7cm/sec. M ore detailsofthe design and from Earth. O ne ofthe technology experim entsplanned implem entationcanbefound in[1],[2],[3]. for the Sojourner m ission w as the reconstruction of the actualpathofthe rover as com pared toit's com m anded The roveris controlled by an Intel-8085 C PU operating at path,so as togive insightforthe design and operation of 2M H z (100KIPS). The on-board m em ory,addressablein futureplanetaryrovers. This paperpresentssomeofthe 16 Kbyte pages,includes 16 Kbyterad-hard PR O M ,176 resultsofthattechnology experim ent. Kbyte EEPROM ,64 Kbyterad-hard R AM and 512 Kbyte R AM . The navigation sensors consist of a rategyro,3 accelerom etersfor sensing the X,Y,and Z axismotion, and 6 w heelencodersforodom etry. A rticulation sensors Abidirectional U H F radio m odem (9600 bits/second) include differential and left and right bogey allow s the vehicletotransm ittelem etry and toreceive potentiom eters. W heel steering and APXS (Alpha- com m ands from earthviathe lander.On-board science Proton X-R ay Spectrom eter) positions are m onitored by instrum entsinclude an Alpha Proton X-Ray 5 potentiom eters. A llmotor currents and the Spectrom eter (APXS) , the W AE (W heel Abrasion tem peratures ofvitalcom ponentsarealso m onitored. Experim ent), and the M AE (Material Adherence Experim ent). The vehicleis pow ered by a 15-wattG aAs solar panel backed up incaseoffailure by a non- rechargeableLithium battery, w hich w as also used for nighttim e APXS operations.

The rover is operated on the basisofafixed local coordinatefram e w ithoriginatthe center of the lander base and the X and Y axes pointed toMartian N orth and East (right-hand rule),respectively(Martian N orthis defined by the Lander sun finder). The vehicle'sX,Y positions arecalculated (at~2 H z rate) by integrating its odom eter(average ofthe six w heelencodercounts) w ith the heading changes produced by the rategyro. D ue to the low processor speed and lack of floating point arithm etic, m illimeter (mm) and Binary Angle M easurem ent (BAM ) areusedasdistance and turn Figure 1:The SojournerR over angle unitsrespectively(1 D eg = 182 BAM or 360 D egs = 65,536 BAM ).Whilemoving,the vehicle m onitored its inclination, articulation, contact sensing, m otor and pow ercurrents,and tem peratures tobesurethey did not exceed limit conditions based on risk level settings. Being too close and heading tow ardlander conditions arealso m onitored. The rover periodically sends a heartbeat signal tothe lander at one vehicle-length intervals. Inthe absence of this com m unication signal, the vehicleis autonom ously backed up halfofitslength and a com m unication retrytakes place. The rover motion is com m anded by one of the follow ing com m ands: Turn, M ove, G o to W aypoint,Find R ock , and PositionAPXS.

The Turn com m and in general causes the vehicleto Figure 2:The roverassem bly change its heading inplace. The four steered w heels are adjusted intotheir appropriate positions, then the The two frontblack and w hite CCD cameras (768 x 484 vehicle w heelsareturned untilthe desired heading, pixels) provide hazard detection and science/operation indicated by integrating the rategyro,ismet.Incasethe im aging. The rearcolorCCD cameraisusedforscience gyroisdisabled, the odom etryisusedtocalculatethe im aging and APXS target verification. A suiteoffive heading changes; if boththe gyro and odom eter are infrared laser stripe projectors, coupled w iththe front disabled,timing isusedinthe calculation. The TurnTo CCD cameras,provide the proximity sensing and hazard com m and causes the vehicletoturntoaspecific detection capabilityforthe vehicle. Thissystem operates heading,w hilethe Turn By com m and causes the vehicle by locating the im age of the laser stripes on a few toturntoarelative heading. The Turn At com m and selected scan lines ofthe cam eraim ages. D eviations of causes the vehicletoturnsoasto pointtoaspecificX,Y the detected locations from the nom inal flat-terrain position. values indicatethat the terrainis uneven. An array of elevation values iscreated from the stripe-cam era The M ove com m and enables the vehicletomovefor a intercepts. P roximity hazards are detected w hen specified distance, using only odom etry and no hazard elevation differences betw een adjacent pointsinthe avoidance. This"blind m ove" isusefulw hen the terrain array exceed a threshold, or w hen the difference isclearly seen by the operator (inim ages from the betw een the highest and low est point inthe array lander) and the m ove is a shortone. The SetSteering exceeds a threshold. O ther hazards include excessive Position param eter of the M ove com m and determines rollorpitch, or excessive articulation of the chassis, or the arcradius ofthe m ove. contact w ith bum p sensorsonthe front or rear of the vehicle. TheGoto W aypoint com m and causes the vehicleto traverse toaspecified X,Ylocation. The vehicledrives forwardadistance of one w heel radius and stops for laser proximity scanning. A terrain height m ap is constructed internallyfrom the information provided by include the obstacle height m ap provided by the the lasers and C C D im agers. Ifan obstacleis detected proximity and hazardavoidance m echanism forevery6.5 on the left,the vehiclewillturnright,and visa versa. A cm oftraverse. flag issetwhich indicates the direction of the turn, and the vehiclewill continue turning by increm ents untila The health check telem etryprovides a snapshot of the hazard-free zone at least as w ide as the vehicleis currentstatus ofthe vehicle. In addition toalmostallof detected by the laserscanning system . Ifthe clearzone the navigation information,the pow ersupplycurrentand iswider than the vehicleturning circle, then the rover voltage status, individual w heel odom eter readings, drives straight ahead far enough tobring the obstacle com m unication errorcounts,device failcounts,m in/max alongside. Then the roverbegins an arctow ardthe goal accelerom etervalues,m otorcurrentvalues,and average point,clearsall m em oryofthe hazardavoidance motor currentsofthe last traversal arereported here. m aneuver,and continues. Ifthe clear zone is narrow er Other rover telem etry datais designed toreport data than the vehicleturning circle(butw iderthan the vehicle) from science, engineering experim ents and rover thena"thread-the-needle" m aneuver isattem pted. This housekeeping utilities. m aneuver centersthe rover on the perpendicular bisector betw een the tw o hazards, and m oves straight InaTurn com m and,the rovercom pletes a turn w hen the ahead along that line until a zone big enough toturn gyro heading isinwithin+/-1.5 degrees of the desired around is detected. O nce such a zone is detected, all heading. In a M ove com m and, the rover com pletes a memoryofthe m aneuveris deleted and the roverbegins m ove once the average six w heel encoder count an arctow ardthe goal.If an obstacleis encountered exceeds the desired encodercount.Partofthe distance prior to detection of a free turning circle,then the rover errorsare due tothe w heelslippage,and they depend on backs straight out tothe point w herethe thread-the- the terrainthe vehicletraverses. needle m aneuverbegan,and the rovercontinues toturn ' untilanother hazard-free zone is detected. Arcs tow ard InGoto W aypoint and R ock Finding com m ands, the the goal arecalculated tothree values: ifthe rover is rover reaches its destination w hen dX * dY < 100 m m 2; already pointed tow ardthe goal (withinasmall dX and dY aredistances from the vehicletoitstarget deadband) then the rover goes straight,ifthe rover position in X and Y respectively. Incasethe rover can heading is outside thatdeadband butless than about1 not get toits destination due toanobstacleatthe radian, then a large-radius turn(about 2 m eters) is destination, the rover declares a successful com m and begun w hich turns tow ardthe goal,and ifthe heading is com pletion w hen itcom es w ithin 500 m m 2 of the target morethan1radian from the goaldirection,then a short destination. The vehicle m onitorsthe progress of the radius turn(about1 m eter)is begun w hich turns tow ard G oTo W aypointand Find R ock com m ands and enforces the goal.Notethanaturninplace m aneuveris notused atimelimit(which is a param eterofthe com m and). here, since that w ouldcausethe rover to becom e trapped in"box canyons" w hereas the presentalgorithm does not. THE ROVER CONTROL W ORKSTATION

The Find R ock com m and isverysimilar tothe G o to The rover isindirectly controlled by hum an operators W aypoint com m and, except that after a hazardis using the R over C ontrol W orkstation (RCW ).The detected at approximatelythe X,Y position of the RCW 'scustom ized graphical user interface software w aypoint,then the rovercentersit's heading betw een the provides toolsfor the operator to generate com m ands edges of the rock using proximity sensing. Ifthe with param eter checking capabilities, and to designate destination coordinates arereached w ithoutany rocks w aypointsina3-Dim age display. A com m and found along the w ay, a spiralsearch is perform ed until sequence w hich com prises m ultiple com m ands is built the rock isfound. G o to W aypoint and Find R ock based on requestsfrom the scientists, vehicle com m ands also containamaximum tim e duration for engineering telem etry,and the end-of-solstereo im ages execution. Ifthat timeis exceeded then the com m and captured by the lander cam eras. The rover 3-Dicon terminates. shownonthe R C W display allow s the operatorto assess traverse abilitybyplacing the icon over a 3-DMartian The Position APXS com m and enables the vehicleto terrainim age set at any position and orientation. The m ove backw ard untilthe APXS sensorhead contactsthe rover'scurrentposition and heading arealso acquired by rock thathas been found oruntilthemaximumallow able matching the icon w iththe rover's physicalposition inthe distance has been reached w ithoutcontactortime-out. stereo im ages. This capabilityallow s the operatortore- initialize the vehicle'strue position and orientation atthe For every uplink com m and, the vehicle sends either an beginning of a sol.InGoto W aypoint designation, the acknow ledge m essage or the telem etrycollected during operator specifies the rover destinations by placing the execution of the com m ands, including any error rover 3-Dcursor at each w aypoint,then clicking the m essages. N avigation telem etryin generalcontains the m ouse toidentifythese destinations. The R C W records time tag, the com m and sequence num ber,the current these w aypoints and generates the G o to W aypoint X,Y and heading values, steering positions, inclination com m ands autom atically. O ther com m ands are and articulation values, m otor currents, tem peratures, generated from operator-specified param eter values, and contactand encoderinformation. In addition,the G o and the com m and sequence fileiscreated. The to W aypoint and R ock Finding telem etry dataalso accuracy of the designation depends on the distance betw een the stereo cam eras, im age resolution, and navigation com m and could notbe m easured. Therefore, hum an designation ability. T he overall accuracy of the inthis analysis,the heading and distance errorsforevery designation w as estimated at about 2 to 3 percent for solare determined from the end-of-solposition estimate cross and dow n ranges,and forheading. m ade by the roveras identified inthe dow nlink telem etry as com pared tothe position ofthe rover as initialized by the driver using the end-of-sol stereo im ages from the THE ROVER NAVIGATION PERFORM ANCE IMP. Thetraversal distance per sol is defined as the ANALYSIS total distances the rover m oved during itstraversals, bothforward(+) and backw ard(-).Thetraversal The rover traversed49sols out of 83 active solson heading persolis defined as the totalchange in heading Mars. Itvisited 16 distinct sites (9rock and 7 soil which the rover integrated during itstraversal, bothleft locations),analyzed them using itson-boardinstrum ents, and rightturns,inthatparticularsol. and captured over500 im ages [4].Theroverhad almost circum navigated the lander inits 100-meter traversal. Extraction of the dow nlink navigation telem etryofevery Since the terrain near the lander w herethe rover had solresultsinthe finalvehicle position and heading,and been deployed is nearly obstacle-free [Figure3],m ostof the totaltraversal distance and totalturn angle per sol. the traversal com m ands used during the first 12 sols The Set VehiclePosition com m and inanuplink werelow -levelcom m ands (i.e.M ove, Turn,and Position com m and ofthe subsequentsoldefines the true position APXS com m ands). The com m ands w ereused and heading ofthe vehicleatthe end ofthe previous sol. thereafter w henever the driver determined thatthe rover This position is determined by the driverthrough the end- might have to negotiatearocky/drop-offterrain. The of-solstereo im ages. The distance betw een the end-ofrover dem onstrated its abilityto negotiaterocky/drop-off sol telem etry position and physicalposition isthe rover terrain on sol24 and sol 33 during the execution ofG o distance error per sol. The difference betw een the end- to W aypoint com m ands. To increase the traversal of-sol telem etry heading and its physicalheading isthe accuracy, the Turn At com m ands w ereusedto adjust heading errorpersol.Figure4showsthe errordistance the rover heading tow ardits destinations beforethe G o as a function oftotaldistance traveled foreach sol. to W aypointorM ove com m ands w ereissued.

Figure 4:ErrorD istance vs.D istance Traveled together withLinear-fitted C urve.

Basedonthe gyro datafrom turn-in-place turns and the Figure 3:O bstacle-free area nearthe ram p odom etry datafrom health checks before and aftereach turn w as perform ed,the turns accounted for by boththe Due torelativelyerratic observed perform ance of the gyro and the odom eter can be com pared. Thereare vehicleinmaintaining it's heading know ledge during the only55turns w hich contain bothgyro-based and mission,the gyro w as deliberatelydisabled after sol49. odom eter-based turninformation. The gyroturnerror Subsequently, allthe turn-in-place turns w erethen percentage is defined as the difference betw een the determined by a virtual heading sensor based on the absolutes of the odom etry-based turn and the gyro- w heel odom eters. The arcing m ovem ent tow ard based turndivided by the absoluteofthe gyro-based destination m echanism inthe G o to W aypointcom m and turn. Figure5showsthe gyroturnerrorsv.s. gyroturn w as autom aticallyreplaced by straight m ovem ent and angles w iththe linear-fitting curve overlaid. The turn turn-in-place m echanism s. errorsaregreater w hen the vehicle m akes right turns; these indicatethatthe gyro w as drifting tothe leftwhile DATA ANALYSIS turning.

Since the true physical position of the vehicle at any m om entw as notknow n untilend-of-solim ages becam e available, the vehicle'sresponse toeverysingle can be reconstructed. O ut of m orethan 1000 stereo im ages, thereare 272 stereo im ages containing the rover tracks w hich can be used for pathreconstruction. Acustom program w as m odified and used totrace allthe rovertracks,thereby determining the XYZ position ofany defined point inthe im age using triangulation together withthe im age's cam era m odel.

Track positions w ere determined and recorded,and the rover physicalpathoverthe entiremission isplotted in Figure 6. The physicalpath,w hich has m any incom plete sections due tolack of stereo im ages containing track information, isdrawnonthe navigation telem etryplot (rover'sinternal-know ledge plot)for com parison. Even withmissing of track data, the com bined draw ing Figure 5:G yroTurnerrorsvs.GyroTurn angles w iththe dem onstrates the accuracy of the rover navigation, and linear-fitting curve overlaid. how the vehicle'sinternal know ledge had perceived its navigation. N otethat for som e sols, the redraw ing of PATH RECONSTRUCTION vehicletracks cannot be done correctly, since som e of the stereo im ages containmultipletracks w ith one The stereo im ages of the rover captured by the lander overlaidthe others. IMP cameras w ereusedto determine the end-of-sol physicalvehicle position and orientation. By extending the use ofthese im ages the true vehicletraversalpaths

Figure 6:show s R over'sPhysicalP ath(darkcolorno straightlines)and R over'sInternal-know ledge Path(lightcolor)ofthe entiremission (Gridsize=1m2) Figure 7:show s R over'sDriver-planned Path(dark & narrow w idthlines,connected designated destinations)and R over's Internal-know ledge Path(widerw idthlines)ofthe entiremission (Gridsize=1m2)

M ostofvehicle end-of-solpositions w erefound tobeto Figure4,itwasthe cross range error com ponent that the rightofitsinternalknow ledge paths as seen inFigure contributed m ost significantlytothe distance error. 6; this observation agrees w iththe gyroleft-drifted Evaluating the cross and dow n range errors behaviordiscussed above. Allthe navigation com m ands mathem atically w ouldbeinappropriatesince therewere wereextracted from the uplink com m and sequences, no lander stereo im ages at the end of everysingle and the navigation planned by the driver for the entire navigation com m and tobeusedin determining the mission isplotted together w iththe rover'sinternal- physical position and orientation of the vehicle. The know ledge plot inFigure7.Thisshowsthe driver- cross and dow n range errors caused by gyrodriftwere expectation destination positions ofroverforeverysingle noticedonEarth during the testing phase,buttherewas navigationalcom m and.These destinations closelymatch no other m icrogyroavailableonthe m arketatthattime the rover'sinternal-know ledge,since itisservoing toit's suitablefor the design and space constraints. W heel internalrepresentation ofthe com m anded path. slippage m ight have contributed insignificantlytothe rover navigation perform ance error since in som e early mission sols(sol 4) w ithstraight m oves, the gyro DISCUSSION heading errorw as low ,resulting intheadistance error.

The overallrover control and navigation design w hich Autonom ous navigation ofthe Sojournerrover,com bined allow s the driver toresetthe vehicle position everysingle with hum an assistance through the R over C ontrol sol had eliminated rover cum ulative navigation errors. Workstation,has proven the rover's capabilitytotraverse The rover 3D -cursor inthe R C W enables the driver to to designated sites for science and engineering m easurethe rover position and orientation accurately, experim ents. The average heading error w as about6.8 and directsthe rovertoits destination.H ow ever,w iththe percent,chiefly due togyroinaccuracy. Testing w iththe inaccuracy ofthe gyro,designation ofrover destinations sam e type of gyro subsequent tothe m ission indicates som etim es becam e cum bersom e and lengthy,especially thatsw itching noise from the D C -D C convertersprobably w hen the APXS w as tobeplaced on a rock. contributed greatlytothe m agnitude ofthe gyrodrift,and that w ithclean pow er the m anufacturers specifications FUTURE MISSIONS for driftareachieved (0.01 deg/sec-root(Hz)).The heading error also influenced the distance error,defined Two currentrover m issions are under developm ent:the as the ratioofthe vehicleerrordistance tothe traversed Athena roverforM ars and the M USES-CN m ission toan distance. The distance error includes the cross and asteroid, w hich isajoint m ission w iththe Japanese dow n range error com ponents,w hich w ereoften difficult space agency ISAS. The Athena rover isthe resultof todisam biguate due tothe com plex natureofthe rover an Announcem entofO pportunityissued by N ASA inthe path and the m any turns involved. Foralmostallsolsin sum m er of 1997. Prof.Stephen Squyres of C ornell Universityisthe Principal Investigator for the science approximateattitudecanbeinferred from the distribution payload forthatrover.TheAthena roverw illbe large (~1 ofpow er from the solarpanelswhich cover the exposed mlong and ~50Kg)and go m uch fartherfrom the lander faces ofthe body ofthe rover. than didSojourner(perhaps Km instead of<10 m ).The M USES-CN rover has been dubbed a nanorover [5], For these futuremissions, the rover w illrequirean since itismuchsmaller than the m icrorover Sojourner improved navigation system , including reliable (~15 cm long,<1Kg),and itwillnothave a lander atall, accelerom eters, an accurategyro, and a sun sensor, butjustfallballistically ontothe asteroidfrom an orbiter. together w ithterrain m apping capability. T hismustbea Thus neither Athena nor the nanorover w ill have the navigation system w ithmuchgreater accuracy,toallow benefitofthe close proximityofalander w hich can be the rover to navigate autonom ouslyfor hundreds of used toprovideafixed observation platform and meters per solw ithmuchless intervention from hum an coordinatefram e. Thus,the Sojourner m ission strategy operators. The rover w illhave toreach its destination of using the lander stereo cam eras toreestablish the preciselyinorder to perform science experim entsas precise position and orientation ofthe roveronce perday com m anded. is notapplicable. ACKNOW LEDGM ENTS Instead, the rovers m ust determine their ow n position and orientation. As w e have seen,the Sojournervehicle losttrack ofit'sorientation relatively quickly due todriftin The authorswish tothank the other m em bersofthe itsrategyro. Inthe case of bothAthena and the R over C ontroland N avigation Team (Jake R . M atijevic, nanorover, som e sortofcelestialnavigation isrequired H enryW.Stone,Andrew H .M ishkin, Jack C . M orrison, tomaintain heading know ledge. (Neither M ars nor the Brian K.C ooper,RichardV.Welch, A llen R .Sirota,and asteroidisthoughttohaveaglobalm agneticfieldwhich Arthur D .Thom pson)w ho created the SojournerC ontrol w ouldbeusefulforheading m easurem ent.) and N avigation subsystem and w ho operated the rover during the m ission. W e w ouldlike to acknow ledge all Athena w ill have a sun sensor,which w illallow the the m em bersoftheMarsPathfinderTeam . direction vector tothe sun to be m easured w ithrespect tothe vehicle coordinatefram e. Acclerom eterswillallow The M icrorover Flight Experim ent (MFEX)is a NASA- m easurem ent of the local gravityvector inthe sam e OACT (Office of Advanced C oncepts and Technology) coordinates. Know ledge of the precise tim e of day w ill activity. T he w ork described inthis publication w as allow the prediction ofw herethe sun shouldbeinthe sky carried out by the Jet Propulsion Laboratory,California (assum ing the latitude and longitude are know n), and Institute of Technology, under a contract w iththe thus allow com putation of the rover's heading inglobal NationalAeronautics and Space Adm inistration (NASA). coordinates. REFERENCES Itisplanned for the nanorover toim age the star field w henever the absoluteorientation inspaceis needed. The cam erawillbe abletoim age starsasdim as about 1. Jacob R . M atijevic, "M arsPathfinder M icrorover - the limit of hum an vision on a darknight,which gives Implem enting a Low C ostPlanetaryMission Experim ent", about4000 starsinthe celestialfield. Any random field- Proceedings ofthe Second IAA InternationalC onference of-view of the cam erawillhave approximatelyvisible7 on Low C ostPlanetary ,John H opkins U niversity,Applied stars expected. Because the angles betw een starscan Physics Laboratory,Laurel,M aryland,April1996. 2. H enryW.Stone, "MarsPathfinder M icrorover: A Low - be accurately m easured insuchanim age, the angles Cost, Low -Pow er Spacecraft", P roceedings of the 1996 and relative brightness betw een any two allow rapid and AIAA Forum on Advanced D evelopm entsin Space precise identification ofthe starsinastarcatalog. Such R obotics,M adison,W I,August1996. acatalogcanbesorted by star brightness,and contain 3. H enryW.Stone, "Design and C ontrol of the the precise angles and ID num bersofthe nearby stars. MESUR/Pathfinder M icrorover", P roceedings of the International C onference on Advanced R obotics, Tokyo, By these techniques, the rover heading know ledge w ill Japan,N ovem ber1993. 4. Andrew .H. Mishkin, J. M orrison, T. N guyen, H . Stone, be m aintained over the long term. However,itis B.C ooper,B.W ilcox, "E xperiences w ith O perations and impractical to continuouslymakestellar observations. Autonom y of the M arsPathfinder M icrorover", Thus Athena w illalso have rategyros (with betterpow er Proceedings of the 1998 IEEE Aerospace C onference, filtering than Sojourner), and the nanorover w illuse March 21-28,Snow m ass atAspen,C olorado. odom etry. O dom etryonanasteroidwith only10 5. Wilcox, B et al, "Nanoroversfor PlanetaryExploration", microgees of surface gravity m ay be suspect,butifthe proc.AIAA R obotics Technology Forum ,M adison W I,pp speed isless than about2 m m per second,then rolling 11-1to 11-6,1-2 Aug 1996. contact w illbe m aintained. At speeds higher than this, the m otion w illbe intermittentballistic hopping. For the M USES-CN m ission,rolling atlow speed w illbe done for precise navigation to nearby science targets,and ballistic hopping w illbeattem pted toreach distant targets. During ballistic hops, star finding w illbe employed to maintainthe know ledge ofthe vehicleattitude. Also,the