Technology Development and Testing forEnhanced Mars Rover Sample Return Operations

�

RichardVolpe Eric Baumgartner PaulSchenker SamadHayatiMailstop198-219 Mailstop82-105 Mailstop125-224 Mailstop180-603

818-354-6328 818-354-4831 818-354-2681 818-354-8273

JetPropulsionLaboratoryCaliforniaInstituteof Technology

Pasadena,California91109Email: firstname.lastname@jpl.nasa.gov

Abstract—ThispaperdescribesseveralJetPropulsionLabo-ratoryresearcheffortsbeingconductedto supportMarssam-plereturnin thecomingdecade.After describingthe2003/05missionscenario,we provide an overview of new technolo-gies emerging from three complementaryresearchefforts:LongRangeScienceRover, SampleReturnRover, andFIDORover. Theresultsshow improvementsin planning,naviga-tion, estimation,sensing,andoperationsfor smallroversop-eratingin Mars-likeenvironments.

TABLE OF CONTENTS

1. INTRODUCTION2. MARS SAMPLE RETURN M ISSION SCENARIO3. LONG RANGE SCIENCE ROVER TECHNOLOGIES4. SAMPLE RETURN ROVER TECHNOLOGIES5. FIDO ROVER TECHNOLOGIES6. SUMMARY7. ACKNOWLEDGMENTS

1. INTRODUCTION

EvenbeforeSojournermadeits first wheeltracksonMarsin1997,it wasanticipatedthatthis roverwouldbeonly thefirstin a seriesof surfaceexplorationspacecrafttargetedfor theplanet. While it will be Sojourner’s flight sparethat driveson Marsin 2002,thenext leapin technicalcapabilityexhib-itedby roverswill bein the2003/05missionset,wheremuchlargerroverswill performrockandsoil samplecollectionforreturnto Earth. Theseroverswill have greaterinnatecapa-bilities, openingthe door for the insertionof new roboticstechnologiesthathave beenin developmentsincethe incep-tion of the Pathfindermissionfive yearsago. Among theseare on-boardstereovision processing,autonomouslander-lessoperations,manipulationandinstrumentpositioningbyarms,precisionnavigation for rover/landerrendezvous,anddistributedgroundoperations.

Of fundamentalimportanceto the incorporationof the newcapabilitieson next-generationrovers is the useof a morecapableelectronics,sensing,andinstrumentationinfrastruc-ture locatedon-boardthe rover. For instance,asSojournerwas being preparedfor flight, JPL was constructinga newprototype,Rocky 7 [16]. Severalkey featureswereaddedtosupportlongtripsawayfrom thelander:adeployablemastto

�0-7803-5846-5/00/$10.00c

�2000IEEE

raisecamerasandtake panoramicimagesof thesurroundingterrain,a shorterarm for sampleacquisitionandinstrumentplacement,anda sunsensorto accuratelydetermineheadingwhile driving. Thesefeaturesweredemonstratedduringfieldtestsin theMojave Desertin 1997,which led directly to theacceptanceof the03/05missions[15].

Theselected03/05missionconcept,however, requiresanen-largedroverthathastheaddedfunctionalityof carryingadrillfor rock sampling,largerwheelsfor enhancedmobility, andasignificantlyupgradedscienceinstrumentsuite(asopposedthe Sojournerrover). Therefore,to supportcontinuedfieldtestswith the selectedscienceteamfor the 03/05 mission,a new Field IntegratedDesignandOperations(FIDO) roverwasconceived, designed,integratedduring a 12-monthpe-riod, anddemonstratedin deserttestsin April 1999[14]. TheFIDO rover reflectsthe currentengineeringsensorsandsci-enceinstrumentsuitethatareplannedfor the03/05mission.While this roverwill continueto actasanoperationstestbedfor missionscientists,it hasa secondfunctionasanintegra-tion testbedfor new technologiesthat continueto be devel-opedby ongoingresearchefforts.

The JPL core robotics technologyprogramhas beensup-porting theseresearchefforts, which include the Rocky 7rover aswell asanotherplatform, the SampleReturnRover(SRR) [13]. Each rover has beendedicatedto increasingautonomyin two respective halvesof the explorationprob-lem: autonomousmotionaway from andbackto the lander.New techniquesusedincludeestimationandvisual localiza-tion,on-boardpathandsequencere-planning,andnaturalandman-madetargetrecognitionandtracking.

This paperdescribesthesetechniques,aswell asthe detailsof the missionscenarioin which they will be used. Sec-tion 2 describestheMarsSampleReturn’sAthenarovermis-sion.Thetechnologydevelopmentsassociatedwith theLongRangeSciencerover taskarediscussedin Section3, whiletheSampleReturnRoverandthedevelopmentof rendezvoustechniquesaredescribedin Section4. TheAthenaterrestrialprototyperover, FIDO, is discussedin Section5, andsum-maryremarksareprovidedin Section6.

Figure 1. Preliminarydrawing of theMarsSampleReturnAthena-Rover.

2. MARS SAMPLE RETURN M ISSIONSCENARIO

Launchingin 2003andagainin 2005,NASA’s MarsSampleReturn(MSR)spacecraftwill placetwo scienceroversonthesurfaceof theplanet.Theseroverswill becarriedto thesur-faceon the top deckof a three-leggedlanderthat is roughly1 m tall and3 m in diameter. This size,substantiallylargerthanpreviouslandersof the1996,1998,and2001missions,makesit possibleto carry a sample-returnMars AscentVe-hicle (MAV) aspart of the landerpayload. The landersizealsoenablestheuseof aroverthatis approximatelytwicethesizeof the Pathfindermissionrover, Sojourner, in every di-mension. (Sojournerwas60 x 40 x 35 cm.) A preliminarydrawing of this “Athena-class”rover is shown in Figure1.

The largersizeof this MSR rover is neededto supportrocksamplingoperationsand to carry the seven scienceinstru-mentsthat make up its payload. Rock samplingis accom-plishedusingacoringdrill, whichrelieson therovermasstoprovide the force behindit in its vertical operatingconfigu-ration. Thescienceinstrumentsareusedto selectthe targetrocksfor sampling,to determinethecompositionof therocksamplesobtained,andto studyrocksin thesurroundingarea.Fourof theinstruments,locatedonafivedegrees-of-freedomarm, requirecloseproximity to the sampleto be measured.Two others,locatedon a 1 m mast,requireonly line of sightto thetarget.Thedrill, itself, is theseventhinstrument.

The rover will begin its missionon the landertop deckbyobtaininga panoramaof thesurroundingterrainfor science,engineering,andpublic outreachpurposes.From theseim-ages,arampdeploymentdirectionwill beselected,aswell asinitial travel routesandgoalsfor thefirst rover traverse.Af-ter rampdeployment,the rover will drive to thesurfaceandbegin navigating the terrain. The maximumdistancedriveneachdaywill be100m,andoftenmuchless,especiallywhentherover is positioningitself for scienceoperations.

During themission,communicationwith theroverwill nom-inally take placetwice perday, relayedby the lander. Eachcommunicationwindow will allow a limited set of imagesanddatato be transmittedto operatorson earth,while newinstructionsareprovided to the rover basedon the previouscommunicationcycle. Typically the rover will receive in-structionsfor thedayin themorning,andtransmittheresultsandstatusat theendof theday.

After obtainingits first setof rocksamples,theroverwill re-turn to thelanderanddepositthemin theMAV. Thismustbeaccomplishedby successfullyaligning with the baseof theramp,driving up its two narrow rails, andaccuratelydetect-ing theproperpositionfor sampletransfer. At this location,theMAV payloaddoorwill beopenedandthesampletransferwill be completedrobustly andautonomously;thermalcon-siderationsrequirethat the payloaddoor be openlessthanthatof thetypical communicationscycle.

Sincemissionconstraintslimit surfaceoperationsto lessthan90days,thesampleacquisitionandreturnto MAV cycle canbeperformedonly threetimesat most. It is likely, however,that eachcycle will seethe rover venturingfartherfrom thelander.

After thelastsamplereturnoperation,theroverwill moveoffthe landerdeckandfar enoughaway from the landerto pre-ventits beingdamagedduringtheMAV lift-of f. This launchis expectedto damagethelandercommunicationsystemandprevent it from acting as a relay for the rover. Therefore,the rover will useauxiliary communicationto an orbiter, toenableit to perform the extendedmissionof exploring thesurface.

Obviously, the complexity andaccuracy of the autonomousoperationsdescribedabove directly influencethe amountofscienceoperationsthat will be performed. For this reason,JPLresearchprojectsaim to introducenew functionalityandfeaturesinto Mars rovers to enablegreatersciencereturnfrom all upcomingMarsrovermissions.Techniquesfor moreautonomousandrobust explorationandreturnto the landerhave beendevelopedandimplementedin field tests.Eachoftheseefforts is describednext.

3. LONG RANGE SCIENCE ROVERTECHNOLOGIES

To improverovernavigation,exploration,andautonomy, theLong RangeScienceRover (LRSR) researchtaskhasbeenimproving therover’sability to navigatethroughtheenviron-ments,while maintainingan accuratesenseof its position.This sectiondescribesadvancesin four pertinentareas:dy-namic sequencegeneration,autonomouspath planning,vi-suallocalization,andstateestimation.All researchwascon-ductedwith theprototyperover, Rocky 7, shown in Figure2.

DynamicSequencePlanning

On-boardplanningwith dynamicsequencegenerationallowsgroundcontrollersto provide muchhigherlevel commands,while increasingtheoptimalityandrobustnessof roveropera-

Figure 2. TheRocky 7 researchprototype.

tionson thesurface.For instance,duringthePathfinderMis-sion, the Sojournerrover [10] wasprovided with extremelydetailedsequencesdaily, fatiguingoperatorswhile alsodis-allowing contingency operationswhentheflow of executionwasnon-nominal.Contraryto this,wehavebeenexperiment-ing with on-boardreplanningthat canchangethe executionof daily activitiesbasedonunanticipatedvariationsin quanti-tiessuchasposition,terrain,power, andtime. To accomplishthis,wehaveusedadynamicon-boardplaningsystemcalledCASPER(ContinuousActivity Scheduling,Planning,Execu-tion, andReplanning)[5], [6].

Figure3 showsanexamplescenarioin mapform,wheredarkorangeshapesrepresentobstaclesknown a priori (e.g. fromlanderdescentimagery). In this case,the initial planfor thetraversewill bring the rover to an unexpectedobstaclenearthefirst goal,representedasalight orangeshadedshape.Cir-cumnavigationaroundthisobstaclewill movetherovercloserto othergoals,triggeringCASPERto recognizethesituationand re-planto visit the closestgoal first. We arecurrentlyevaluatingthisandsimilarscenariosexperimentally.

AutonomousPathPlanning

For the longertraversesrequiredof upcomingmissions,au-tonomouspathplanningis desirablesinceoperatorswill notbe able to seethree-dimensionalterrain featuresout to themoredistantgoallocations.WhereasSojournerdrovea totalof 84m duringits entiremission,theMSRAthenaroverwillbecapableof driving this distancein a singleday. However,stereoimageryof the terrainprovidedto operatorswill onlyhave an envelopeof 20 m at bestresolution.Therefore,thepathplanningadvancesdescribedherewill allow therovertobeits own operator. It canimagetheterrainfrom periscopic

Re-planned path

Original planned path

Figure 3. After encounteringapreviouslyunknownobstacleshown in light orange,CASPERreplansthe sequenceof targets.Bluecrossesarethegoallocations,anddarkorangeshapesareobstaclesknown apriori.

cameras,selecta paththroughthe terrainto the edgeof theeffective stereorange,andrepeatthe processuntil the goalis achieved. A representative exampleof a partialpanoramaand the resultingelevation mapof the terrainareshown inFigure4.

Rocky 7 usesa localsensor-basedpathplannercalledRover-Bug [8], [9]. This algorithmwasdevelopedfor vehiclesthathave limits on sensorrange,field of view, and processing.The two main modesof operationare motion-to-goalandboundary-following, which areusedto provide global con-vergence.

Rover-Bug works by using the local elevation map to con-structa mapof convex hulls aroundall obstaclesin thesens-ing range.Thesehullsarethenmergedandgrown to providea configurationspacerepresentationof thesensedterrain. Atangentgraphis constructedto determineif thereis an un-obstructedpath to the envelopeof the sensedregion in thedirectionof thegoal. If oneexists,it is followed,andthepro-cessis repeatedat theendof thepathsegmentto thesensoryenvelope.

If a free path doesnot exist, stereoimagesareobtainedtothemostpromisingsideof thecurrentview, andtheprocessis repeated.In somecases,the free pathplacesthe rover attheedgeof thesensoryenvelopebut still obstructedfrom thegoalby anobstacle.In this case,therover will begin to useits body-mountedcamerasto reactivelyboundaryfollow untilthereis aclearpathto thenext goal.

Figure 5 shows experimentaldata obtainedfrom Rocky 7while usingthisalgorithmto traverseourMarsYardtestarea.The startpositionis in the lower left corner, 21 m from thegoalin theupperright. Foursetsof sensorydataareshown as

Figure 4. Stepsof on-boardterrainsensing:panoramicmosaicview from rover maststereoimager, compositerangemapextractedfrom stereoviews,andelevationmapcreatedfrom rangedata.

Figure 5. Experimentalresultsfrom a multi-steprun usingRover-Bug in theJPLMarsYard.

orangewedgesalongthepath.Projectedon to thesedatasetsarethepink convex hull representationsof thesensedobsta-cles. The greenandblue lines passingthroughthe obstaclefield are the plannedandexecutedpaths,respectively. Thegapbetweendatasetsis dueto alackof mergingof datafromboth the mastandbody-mountedcameras,and is currentlybeingcorrected.Thesuddenchangesin thepathdirectionatthefarsideof eachwedgedoesnotindicatedactualrovermo-tion; ratherit is anartifactof theestimatedroverpositionthatis updatedby localizationat this point in the traverse. Thislocalizationis discussednext.

VisualLocalization

Visual localization usesthe sameterrain imagery as pathplanning,but for thepurposeof monitoringtheapparentmo-tion of three-dimensionalgroundfeaturesafter therover hascompleteda move. In this way, the on-boardpositionesti-mateof the rover can be updatedto compensatefor errorscausedby wheelslippageor rock bumping. On Pathfinder,thislocalizationfunctionalitywasperformedmanuallybyop-eratorsviewing Sojournerfrom thefixedpositionlandercam-eras,restrictingtheupdateto onceadayandpermittingoper-ationsonly within thestereoenvelopeof the lander. In con-trast,the terrain-basedlocalizationdescribedherehasappli-cationto many forms of landerlessoperations:incrementallong traverses,local operationswithin rangeof a prior stereopanorama,localizationin descentimagery, andclosedchainrovermoveswith estimatesmoothing.

Goal position

Localization target

Figure 6. Exampleof automatictargetselectionfor local-ization.

Thetechniquereliesonobtainingastereoelevationmapfromtheinitial roverposition,asshown in Figure4. This imageryis automaticallyanalyzedbasedontherangedataquality, andthequality expectedto beseenat thegoalpoint specifiedbypathplanning[11]. Typically, therewill bea prominentrockbetweenthe two locations. From the analysis,it will be se-lectedasthelocalizationtarget,andviewedby theroverafterreachingthe endof the local pathsegment. Figure6 showsanexampleof theautomatictargetselection,with four imagesconcatenatedto provideareasonablenumberof potentialtar-gets.

To match the two stereo views of the terrain, a multi-resolutionsearchtechniqueis usedto provide the bestesti-mateof thedisplacementof theoriginal elevationmapfromthe final one[12]. Typical resultsprovide positionerror es-timatesthat are1% of the distancetraveled. However, thesearchusestheon-boardestimateof therover positionasitsstartingpoint,somorereliableresultsareobtainedif theini-tial estimateis moreaccurate.Therefore,to obtainbettercon-tinuouspositionestimates,we have beendevelopinga newestimationtechnique,discussednext.

0 2 4 6 8 10−10

0

10

20

ξ1

(deg

) ESTSIM

0 2 4 6 8 10−10

−5

0

5

10

ξ2

(deg

) ESTSIM

0 2 4 6 8 10−10

−5

0

5

10

t (s)

ξ3

(deg

) ESTSIM

0 2 4 6 8 10−20

−10

0

10

ξ4

(deg

) ESTSIM

0 2 4 6 8 10−10

−5

0

5

ξ5

(deg

) ESTSIM

0 2 4 6 8 10−5

0

5

10

t (s)

ξ6

(deg

) ESTSIM

Figure 7. Simulationresultsshowing truevaluesandesti-matesfor wheelcontactpositions,wherethe left/rightthree wheelsanglesare shown by the left/right sidegraphs.

KinematicStateEstimation

In addition to periodic visual localizationof the rover, wehavedevelopedreal-timepositionandheadingestimationus-ing all othersensorson thevehicle:angularrate,accelerom-eter, sunsensor, wheelanglerates,andmobility systemlink-age(rocker-bogey) configuration.This techniquemovesfarbeyondthesimpledeadreckoningof Sojournerandimprovesuponour previous advancesin positionestimationwith sunsensing[15]. The resultsaid navigation during pathexecu-tion, provide betterinput to localization,andreplacebothinvisually featurelessterrain(e.g.,sanddunes)or in thecaseofvisualsensingfailure.

Theestimatorusesa Kalmanfilter framework, with thepro-cessmodelchosensothattheinertial sensordataareusedasaninput to drive theprocessequation.Theprocedureavoidsthe difficulty of modelingthedetailedprocessdynamics[1]by exploiting the ability of the Kalmanfilter to performtheappropriateleast-squaresaveragingof theactionof eachkine-maticchainin therover. Theseforwardkinematicchainshavevelocitycomponentsdefinedby thesequenceof links joiningthe rover frame to eachestimatedwheelcontactpoint, anda componentgiven by the slip betweenthe wheel and theground. The deterministiccomponentof the slip is usedtocapturetheeffectsof a known steeringactionor a known av-erageslip rateoverdifferentkindsof terrain.

Bothsimulationandexperimentalresultswith thisestimationtechniquehavebeenconducted.Figures7showstheability oftheestimatorto correctlytrack thewheelcontactangleoveranundulatingterrain.Thesimulationis particularlyvaluablesincegroundtruthcanbeknown exactly. For instance,usingsimpleintegrationof thewheelvelocity resultsin a 2% errorin measuredposition,while theestimatorreducestheerrorbyanorderof magnitude.

Experimentalresultsfurthersupporttheseconclusions.Fig-ure8 shows theresultsfrom Rocky 7 driving over anobsta-cle on the right sideof the vehicleonly. While the left side

0 5 10 15 20 25−10

−5

0

5

ξ1

(de

g)

0 5 10 15 20 25−2

−1

0

1

ξ2

(de

g)

0 5 10 15 20 25−1

0

1

2

t (s)

ξ3

(de

g)

0 5 10 15 20 25−20

−10

0

10

ξ4

(de

g)

0 5 10 15 20 25−10

0

10

20

ξ5

(de

g)

0 5 10 15 20 25−20

−10

0

10

t (s)

ξ6

(de

g)

Figure 8. Experimentalresultsfrom Rocky 7 showing theestimatesfor wheelcontactangles.

Figure 9. TheSampleReturnRover.

wheel contactpoints changeonly slightly, the much largerchangesin the right sidearecorrectlydetermined.Note, inthefinal configuration,thebogey wheelswerestill on theob-stacle,while thesteeringwheelhadcompletelytraversedtheobstacle. The estimatedwheel contactangleswere within5 degreesof thetruevaluesmeasuredindependently. Furtherexperimentsarein progress,aswell asintegrationof visualpositionestimation,asdescribedin thenext section.

4. SAMPLE RETURN ROVER TECHNOLOGIES

In 1997,JPLbegandevelopmentof theSampleReturnRover(SRR),a small, lightweight rover that investigatedfocusedtechnologyadvancesin theareasof rover-to-roverandrover-to-landerrendezvous. SRR, shown in Figure 9, is a 7 kgrover with a 4-wheel rocker mobility systemthat is capa-ble of traversingover obstaclesup to 15 cm in height. Therover includes4-wheelsteeringandcarriesa threedegrees-of-freedommanipulatorarmwith a1degree-of-freedomgrip-per. This robot arm is usedfor panoramicimaging alongwith samplepick-up and transfer. The rover also includesaposablerockerjoint for variablegroundclearanceandroverreconfigurationin difficult terrain. SRR carriesa PC104+derived electronicssystem,including a 300 MHz AMD K6processor, motioncontrolI/O boards(D/A andencoderread-erswith closed-loopcontrolrealizedin software),A/D board,PCIcolor framegrabbers,andawirelessEthernet.

Theoriginal conceptfor SRRderivedfrom thepreviousver-sion of the MSR missionwherean Athena-classrover tra-

versedlongdistances,stoppedatinterestingsciencesites,andacquired� andstoredrock andsoil samples.During a secondmission,asmall,lightweightroverwouldlandontheMartiansurfaceandrendezvouswith thesciencerover to retrieve thesamplecollectedduring the primary sciencemission. Sucha rover-to-rover rendezvous was demonstratedby the SRRtechnologyteamduring thesummerof 1998[13]. This ren-dezvouswas accomplishedusing an RF beacontransmitterand receiver pair for non-line-of-sitenavigation to the sci-encerover, visual trackingof thestaticsciencerover duringline-of-sightnavigation,determinationof rover-to-roverposeusingman-madefeatureslocatedon the sciencerover, and,terminalguidanceto thesamplecachecontainerandpickupof thiscontainerusingtheon-boardmanipulatorarm.

In 1999, the SRR task turned its attentionto the rover-to-landerrendezvousproblemin supportof the MSR missionandthe requirementthat the Athenarover return its cachedsamplesto the landerand the awaiting MAV, as describedin Section2. Someof the technologydevelopmentsestab-lishedduringtherover-to-roverrendezvousapplicationweretransferredto the rover-to-landerrendezvousproblem. Therequirementsof theAthenarover mission,however, necessi-tatedthe developmentof new techniquesfor the robust andaccuratenavigation of SRRwith respectto the landersuchthat autonomouslanderacquisition,landerrendezvous,andrampclimbingarepossiblewith minimal, if any, groundsup-port. Figure10depictsthescenarioassociatedwith therover-to-landerrendezvousproblemin termsof themulti-phaseop-erationsandtechnologiesusedduringthereturnto thelander.The rover-to-landerrendezvousproblemis divided into thefollowing phases:

� Longdistancevisualtrackingusinglandertexturefeaturesderivedin thewaveletspace

� Multi-point tracking of lander featuresfor headingandrangeestimationof theroverrelative to thelander

� Ramplocationdeterminationusinglanderfeatures� Ramp recognition using cooperative ramp featuresfor

headingand rangeestimationof the rover relative to thebottomof theramps

Thelong-distancenavigationof SRRrelative to thelanderisaccomplishedwith anovelwavelet-baseddetectionalgorithmfor thelong-rangevisualacquisitionof a landerfrom greaterthan 100 m using a single, black-and-whiterover imagingsystem(20 degreefield of view) [4]. This informationen-ablesautonomouscorrectionthe rover headingwith respectto the lander, andto guidanceof SRRto within 25 m of thelander. Sucha navigation sequenceis shown in Figure11.Within 25 metersof the lander, a visual techniquethat takesadvantageof the known geometryof the landerstructureisusedto trackmultiple features(e.g.,landerleg struts,landerdeck,etc.) to determinetheposeof the rover relative to thelander. Preliminaryresultsindicatethatprecisionontheorderof 50cmin rangeand1 to 2 degreesin orientationis possibleusingthisapproach.

Likewise,theknown rampgeometryallowsfor thevisualac-quisition of the landerrampsandthe relative positioningoftherover relative to theramps.Thesetechniquescombinetoproducea navigation strategy for the autonomousguidance

of the rover from 25 m from the landerto 5 m in front ofthe ramps.Finally, visual acquisitionof the landerrampsisaccomplishedusingcooperativemarkingsfrom which rover-to-ramppositionandheadinginformation is obtained. Thesuccessive visual acquisitionof the landerrampsbringstherover from 5 metersto within 5 cm of thebottomof the lan-der ramps.Many successfulexperimentshave beenaccom-plishedin bothlaboratoryandoutdoorsettings.Theseresultsindicatethattherovercanreliableandrobustlynavigateto thebottomof therampswith anabsoluteprecisionof 1 cmin lat-eralandlongitudinaloffsetandlessthan1 degreeorientationerrorwith respectto theramps.As such,thecombinationoftheserovernavigationtechniquesleadsto a single-commandautonomoussequenceassociatedwith thereturnto thelanderandtheregressof theroverupthelanderrampsto depositthesamplecachein theMAV.

Finally, theSRRtaskhas,over thepasttwo years,developedanalternative form of rover stateestimationfor theaccurateandreliabledeterminationof rover positionandorientationrelativeto afixedreferenceframe.Thistechniqueusesanex-tendedKalmanfilter framework basedonthework describedin [3]. Within theSRRdevelopment,theregistrationof suc-cessive rangemapsgeneratedby therover’s forward-lookinghazardavoidancecamerasareutilizedtodeterminetheframe-to-frametranslationandrotationof the rover. This informa-tion is combinedwith deadreckonedestimatesof therover’stranslationandrotationto producean optimizeddetermina-tion of therover pose.This work is describedin [7] and[2]andillustratedin Figure12 for a 6+ m traversewithin a soft-soil, rock-filled indoorsandpit.

5. FIDO ROVER TECHNOLOGIES

As describedin Section2, the2003/05MSR’s Athenarovermissionrepresentsa significantincreasein complexity overtherecentSojournerrovermission.TheAthenarovercarriesseven sciencepayloadelements,including a multi-spectralimagingsystem,amicroimager, asampleacquisitionsystem,andfour differentspectrometersascomparedto Sojourner’ssinglescienceinstrument,theAlpha ProtonX-ray spectrom-eter. In addition,the Athenarover hasover eight timesthevolumeand six times the massof the Sojournerrover. Assuch,theAthenarover representsa truesciencecraftandful-fills theneedfor a roboticfield geologistonMars.

To facilitatethesuccessfuloperationof theAthenaroverdur-ing the 2003/05Mars SampleReturnmissions,a terrestrialprototypeof theAthenaroveris beingusedby theAthenasci-enceteamfor scienceandengineeringtestingassociatedwithflight missionoperations. The developmentof this roboticvehicle,known as FIDO (for Field Integrated,Design,andOperations)rover began in early 1998 with the conceptualdesignof thevehicleandculminatedninemonthslaterwiththefull-scaleintegrationof theroverandits sciencepayload[14]. In April 1999(lessthan14monthssincetheinitial paperdesigns),theFIDO roverwassuccessfullyoperatedattheSil-ver Lake field testsiteoutsideBaker in California’s Mojavedesert. Figure13 shows FIDO operatingduring this desertfield test.

FIDO’s mobility subsystemconsistsof a 6-wheel rocker-

Figure 10. Theoperationsscenarioassociatedwith thereturnto thelander.

Figure 11. Wavelet-basedlanderdetectionandnavigation.

Kalman FilteredDead Reckoned

0 10 20 30 40 50 60 70 800

0.5

1

1.5

2

2.5

3Position Est. Error wrt. Ground Truth, K.F. vs. D.R.

Time (Navigation Iterations)

Pos

ition

Err

or (

m)

Figure 12. Stateestimationfor roverposedetermination.

Figure 13. TheFIDO roverduringtheSilverLakefield trial.

bogiesuspensionsystemthathasbeenscaledup by a factorof 20/13 from the Sojournerdesign. This suspensionsys-temallows for thesafetraverseoverobstaclesup to 30cminheight. Eachwheel is independentlydriven andsteeredus-ing a Sojourner-derivedactuationandencodersystem. Thetop groundspeedof the vehicle is 9 cm/sec. Approximaterover dimensionsare1 m in length,0.8m in width, 0.5m inheight,and0.23m groundclearance.Therovercarriesafourdegrees-of-freedomdeployablemastthat stands1.94 m offthegroundsurfaceat full extent.Thismastprovidesthenec-essarypan-and-tiltcontrol for panoramicimagingandpointspectroscopy. FIDO alsocarriesa four degrees-of-freedominstrumentarmthatis usedto placethein situsuiteof instru-mentson rockandsoil targets.

TheFIDO electronicsaresimilar in natureto theSRRelec-tronicswith theCPUbeinga 80586AMD processorrunningat a 133 MHz clock speed. The rover usesa PC104-basedplatform for all I/O functions, including a motion controlsystem(D/A and encoderreaderswith closed-loopcontrolin software) that can control up to 30 actuatorssimultane-ously, two monochromaticandonecolor framegrabbers,dig-ital I/O boards,A/D boards,low-passfilter andanalogmul-tiplexerboards,andawirelessEthernet.Engineeringsensorsinclude front and rear stereohazardavoidancecamerasys-tems,aninertialnavigationsystem,andasunsensorfor abso-luteheadingdetermination.A differentialGPSunit is alsoin-tegratedwithin theroverelectronicsfor ground-truthingpur-posesonly.

Figure 14. Theminiaturecoredrill acquiringacoresamplefrom a carbonaterockatSilverLake

ThesciencepayloadonFIDOis analogousto theAthenapay-load. In particular, the remotesensingsuite locatedon theFIDO mast includesa multi-spectral,narrow field-of-viewPancam stereo imaging system; a monochromatic,widerfield-of-view Navcamstereoimagingsystem;andtheopticsfor a near-IR point spectrometerthatoperatesin the1200to2500 nm wavelengthregion. The multi-spectralcapabilityassociatedwith the Pancamsystemis realizedusinga Liq-uid CrystalTunableFilter (LCTF) that is tunedto the threenear-IR wavelengthsof 650, 750 and 850 nm. The in situinstrumentsuiteattachedto theend-effectorof theFIDO in-strumentarm consistsof a color microimageranda Moess-bauerspectrometerthatareusedto determinetheiron contentof targetrocks.A miniaturecoredrill system,body-mountedto therover, providesthecapabilityto acquireandcacherockandsoil samples.All of theseinstruments,with theexceptionof the near-IR point spectrometer(the flight missionusesaminiaturethermalemissionspectrometerin themid-IR wave-length region), arebreadboardsof the Athenaflight instru-ments.

In total,thescienceinstrumentandengineeringsensingsuitesand the resultingFIDO rover systemrepresenta terrestrialanalogof theAthenarover thatcanbeusedto testandvali-datetheAthenamissionscenarioandassociatedengineeringfunctions.As such,theAthenascienceteamledby ProfessorSteven Squyres,AthenaPI, andProfessorRaymondArvid-son, AthenaCo-I, has worked with the FIDO engineeringteamsinceMarch1999to performroveroperationstestinginsupportof theAthenarovermission.In particular, thedesertfield trial at theSilverLake testsiterepresentedthefirst everdemonstrationandvalidationof thesampleacquisitionphaseof theAthenarovermissionthroughtheidentificationof tar-getrocks,approachto thetargetrock,placementof theminia-turecoredrill over the target rock, successfulacquisitionofa core sampleusing the miniaturecore drill, and return ofthesesamplesto a simulatedlandingsite. Figure14 showstheFIDO rover andassociatedcoredrill duringthesuccess-ful acquisitionof acoresamplefrom acarbonaterock. Futurefield trials in 2000and2001will focusonflight-likeroverop-erations,with therover andscienceteamsbeingsequesteredat JPLwhile therover is locatedat a remotetestsite. Duringsuch“blind” field trials,thefull Athenamissionscenariowillbethefurthervalidated,includingsampleacquisitionandre-turn to the landeraswell as long-rangescienceexplorationanddiscovery.

6. SUMMARY

This paperhasreviewed several recentresearchefforts thatsupportupcomingMSR missionscenarios.Threeresearchtasks,LRSR,SRR,andFIDO,havebeendevelopingandtest-ing new capabilitiesin prototyperover platforms.LRSRre-searchwith Rocky 7hasdevelopedfournew techniquestoen-hancethefunctionalityof autonomouslong-rangeMarsrovernavigation: intelligent sequencing,sensorconstrainedpathplanning,naturalterrainvisual localization,andKalmanFil-ter stateestimation. SRR hasdevelopeda return-to-landercapabilitywith visual trackingat variousranges,usingtech-niquessuitableto thoseranges.It hasalsodisplayedvisualodometrytechniques.Finally, FIDO representstheculmina-tion of thetwo corerobotictechnologyprogramssinceits de-velopmentis basedon the experiencesgainedwithin LRSRandSRR.As aresult,it hasbecomeastandardintegrationandtestsystemfor thevalidationof rover navigationandcontrolstrategies.

7. ACKNOWLEDGMENTS

Largesystemslike theserequirethesupportof many peoplebeyondtheauthors.We would like to recognizethe follow-ing individualsfor their invaluablecontribution to the workdescribedin this paper: the LRSR teamincludesJ. (Bob)Balaram,Clark Olson,SharonLaubach,TaraEstlin,RichardPetras,DarrenMutz, Greg Rabideau,and Mark Maimone;theSRRteamincludesHrandAghazarian,TerryHuntsberger,YangCheng,MikeGarrett,andLeeMagnone;theFIDOteamincludesHrandAghazarian,TerryHuntsberger, Anthony Lai,RandyLindemann,Paul Backes,Brett Kennedy, Lisa Reid,MikeGarrett,andLeeMagnone.

The researchdescribedin this paperwascarriedout by theJetPropulsionLaboratory, CaliforniaInstituteof Technology,underacontractwith theNationalAeronauticsandSpaceAd-ministration. Referenceherein to any specificcommercialproduct,process,or serviceby tradename,trademark,man-ufacturer, or otherwise,doesnot constituteor imply its en-dorsementby theUnitedStatesGovernmentor theJetPropul-sionLaboratory, CaliforniaInstituteof Technology.

REFERENCES

[1] J. Balaram. Kinematic State Estimation for a MarsRover. Robotica, Special Issue on Intelligent Au-tonomousVehicles, Acceptedfor publication,1999.

[2] E. T. Baumgartner, P. C. Leger, P. S.Schenker, andT. L.Huntsberger. SensorFusedNavigationandManipulationfrom a PlanetaryRover. In SensorFusionand Decen-tralizedControl in RoboticSystems,SPIEProceedings3523, Boston,November, 1998.

[3] E. T. Baumgartnerand S. B. Skaar. An AutonomousVision-BasedMobile Robot. IEEE Transactionson Au-tomaticControl, 39(3):493–502,March1994.

[4] J-K. Changand T. L. Huntsberger. Dynamic MotionAnalysisusingWavelet Flow SurfaceImages. PatternRecognitionLetters, 20(4):383–393,1999.

[5] S.Chien,R. Knight, R. Sherwood,andG. Rabideau.In-tegratedPlanningandExecutionfor AutonomousSpace-

craft. In IEEEAerospaceConference, AspenCO,March1999.

[6] T. Estlin, G. Rabideau,D. Mutz, andS. Chien. UsingContinuousPlanningTechniquesto CoordinateMultipleRovers.In IJCAIWorkshoponSchedulingandPlanning,Stockholm,Sweden,August1999.

[7] B. D. Hoffman, E. T. Baumgartner, T. L. Huntsberger,andP. S.Schenker. ImprovedRover StateEstimationinChallengingTerrain.AutonomousRobots, 6(2),1999.

[8] S. Laubach. Theoryand Experimentsin AutonomousSensor-BasedMotion Planning with Applications forFlight PlanetaryMicrorovers. PhDthesis,CaliforniaIn-stituteof Technology, May 1999.

[9] S. Laubachand J. Burdick. An AutonomousSensor-BasedPath-Plannerfor PlanetaryMicrorovers. In IEEEInternationalConferenceon Roboticsand Automation,DetroitMI, 1999.

[10] J. Matijevic et al. Characterizationof theMartianSur-faceDepositsby the Mars PathfinderRover, Sojourner.Science, 278:1765–1768,December5 1997.

[11] C. Olson. Subpixel LocalizationandUncertaintyEsti-mationUsing Occupancy Grids. In IEEE InternationalConferenceon Roboticsand Automation, pages1987–1992,Detroit,Michigan,May 1999.

[12] C. OlsonandL. Matthies. Maximum-likelihoodRoverLocalizationby MatchingRangeMaps. In IEEE Inter-nationalConferenceonRoboticsandAutomation, pages272–277,Leuven,Belgium,May 1998.

[13] P. S. Schenker et al. New PlanetaryRovers for LongRangeMars ScienceandSampleReturn. In IntelligentRobotsand ComputerVision XVII, SPIE Proceedings3522, Boston,November, 1998.

[14] P. S. Schenker et al. FIDO Rover and Long-RangeAutonomousMars Science. In Intelligent RobotsandComputerVisionXVIII, SPIEProceedings3837, Boston,September, 1999.

[15] R. Volpe.NavigationResultsfrom DesertFieldTestsoftheRocky 7 MarsRover Prototype.InternationalJour-nal of RoboticsResearch, 18(7),1999.

[16] R.Volpeetal. Rocky 7: A Next GenerationMarsRoverPrototype. Journal of AdvancedRobotics, 11(4):341–358,1997.

Richard Volpe, Ph.D., is the Principal Investigatorfor theLongRangeScienceRoverResearchProject,andtheSystemTechnologistfor the Athena-Rover samplecollection robotfor the 2003Mars SampleReturnProject. His researchin-terestsincludereal-timesensor-basedcontrol, robot design,softwarearchitectures,pathplanning,andcomputervision.

Richardreceived his M.S. (1986)and Ph.D. (1990) in Ap-plied Physicsfrom Carnegie Mellon University, where hewasa US Air ForceLaboratoryGraduateFellow. His the-sisresearchconcentratedon real-timeforceandimpactcon-trol of robotic manipulators.SinceDecember1990,he hasbeenat theJetPropulsionLaboratory, CaliforniaInstituteofTechnology, wherehe is a SeniorMemberof the TechnicalStaff. Until 1994,hewasamemberof theRemoteSurfaceIn-spectionProject,investigatingsensor-basedcontrol technol-ogy for teleroboticinspectionof theInternationalSpaceSta-tion. Startingin 1994,he led thedevelopmentof Rocky 7, anext generationmobilerobotprototypefor extended-traversesamplingmissionson Mars. In 1997,he received a NASAExceptionalAchievementAwardfor thiswork, whichhasledto thedesignconceptsfor the2003Marsrovermission.

Eric T. Baumgartner, Ph.D., is a group leaderin the Me-chanicalandRoboticsTechnologyGroupanda seniormem-ber of engineeringstaff in the ScienceandTechnologyDe-velopmentSectionat NASA’s JetPropulsionLaboratoryinPasadena,CA. At JPL,heservesin asystemsengineeringca-pacityfor thedevelopmentof advancedplanetaryroversandalsocontributesto technologydevelopmentsin the areasofrobotic sensingand control. Prior to his tenureat JPL, hewasan AssistantProfessorin the MechanicalEngineering-EngineeringMechanicsDepartmentat MichiganTechnolog-ical University in Houghton,MI. He haspublishedover 30articlesin the areaof robotic, controls,andstateestimationandis activein theSPIEandASME.HereceivedhisB.S.de-greein AerospaceEngineeringfrom theUniversityof NotreDame,the M.S. degreein AerospaceEngineeringfrom theUniversityof Cincinnatiin 1990,andthePh.D.in Mechani-calEngineeringfrom theUniversityof NotreDamein 1993.

Paul S. Schenker, Ph.D., is Supervisorof the Mechani-cal andRoboticsTechnologiesGroup,MechanicalSystemsEngineeringandResearchDivision, JetPropulsionLabora-tory. His current work emphasizesplanetaryrover devel-opmentandrobotic samplingtechnologies;he is taskman-agerfor NASA/JPL’s SampleReturnRover (SRR)andEx-ploration TechnologyRover Design, Integration, and FieldTest R&D efforts (”ET Rover/FIDO” – a Field IntegratedDesign& Operationsrover prototypesupportingthe NASAMars ’03-’05 samplereturn missions, and related terres-trial missionsimulations),as well as new NASA tasksonuseof cooperatingroboticassets/roversfor Marsexplorationand future humanhabitation. Schenker also recently ledJPL’s PlanetaryDexterousManipulatorsR&D underNASAfunding, work that proto- typed a robotic sampling con-cept which flies on the NASA Mars Polar Landermissionnow in route to the red planet. Schenker’s other recentroboticsR&D activities includea role asfoundingco-PI forNASA/MicroDexterity SystemsInc. developmentof aRobotAssistedMicrosurgeryhigh-dexterity tele-operativeworksta-tion andalongerstandinginvolvementin varioustele-robotictechnologyand systemdevelopmentsfor orbital servicingandautonomousroboticexploration. Schenker is a memberof AAAI, IEEE, and SPIE;he is a Fellow and1999Presi-dentof thelast. Schenker is widely active in externaltechni-cal meetings,publications,and university collaborationsinthe areasof roboticsand machineperception,having con-tributedabout100 archival articlesto same. Dr. Schenkerreceivedhis B.S. in EngineeringPhysicsfrom Cornell Uni-versity, andcompletedhisM.S.,Ph.D. andpostdoctoralstud-ies in ElectricalEngineeringat PurdueUniversity. Prior tojoining JPL/Caltechin 1984,Schenker waswith theElectri-calSciencesfaculty, BrownUniversity, andlatertheResearchSectionChieffor Signal& ImageProcessing,Honeywell Inc.

Samad Hayati received his M.S. andPh.D. in MechanicalEngineeringwith a specialtyin controlsfrom theUniversityof CaliforniaatBerkeley in 1972,and1976,respectively. Hejoined the JetPropulsionLaboratoryin 1979andworked intheguidanceandcontrolof theJupiterorbiterGalileospace-craft, currently orbiting Jupiter. From 1983-1999he per-formedresearchin roboticsat JPL.He wasalsoa lecturerofroboticsat theCalifornia Instituteof Technologyfrom 1986to 1988.He haspublishednumerousconferenceandjournalpapersandholdstwo US patentsrelatedto roboticscontrol.His pioneeringwork in robotcalibrationwasusedto developtechniquesto utilize manipulatorsasan aid in brain neuro-surgery at the Long BeachMemorial Hospital in Californiain 1986.His mostrecentresearcheffortswerein thedevelop-mentof long rangescienceroverandrendezvousandsampleretrieval technologiesfor NASA’splannedmissionsto Mars.

Currently, Samadis the managerof RoboticsandMars Ex-plorationTechnologyProgramsatJPL.