nist -利用人工智能和自主實驗室推動美國材料與制造創(chuàng)新 Driving U.S. Innovation in Materials and Manufacturing using AI and Autonomous Labs

NISTSpecialPublicationNISTSP1320

DrivingU.S.InnovationinMaterialsand

ManufacturingusingAIandAutonomousLabs

BuildingaNationalCenterandTestbedforAutonomousMaterialsScience

HowieJoress ZacharyTrauttAustinMcDannald BrianDeCostAaronGiladKusneFrancescaTavazza

Thispublicationisavailablefreeofchargefrom:

/10.6028/NIST.SP.1320

NISTSpecialPublicationNISTSP1320

DrivingU.S.InnovationinMaterialsandManufacturingusingAIandAutonomousLabs

BuildingaNationalCenterandTestbedforAutonomousMaterialsScience

HowieJoress ZacharyTrauttAustinMcDannald BrianDeCostAaronGiladKusneFrancescaTavazza

DataandArtificialIntelligenceDrivenMaterialsScienceGroupMaterialMeasurementLaboratory

Thispublicationisavailablefreeofchargefrom:

/10.6028/NIST.SP.1320

August2024

U.S.DepartmentofCommerce

GinaM.Raimondo,Secretary

NationalInstituteofStandardsandTechnology

LaurieE.Locascio,NISTDirectorandUnderSecretaryofCommerceforStandardsandTechnology

Certainequipment,instruments,software,ormaterials,commercialornon-commercial,areidentifiedinthis

paperinordertospecifytheexperimentalprocedureadequately.Suchidentificationdoesnotimply

recommendationorendorsementofanyproductorservicebyNIST,nordoesitimplythatthematerialsorequipmentidentifiedarenecessarilythebestavailableforthepurpose.

NISTTechnicalSeriesPolicies

Copyright,Use,andLicensingStatements

NISTTechnicalSeriesPublicationIdentifierSyntax

PublicationHistory

ApprovedbytheNISTEditorialReviewBoardon2024-07-10

HowtocitethisNISTTechnicalSeriesPublication:

JoressH,TrauttZ,McDannaldA,DeCostB,KusneAG,TavazzaF(2024)DrivingU.S.InnovationinMaterialsandManufacturingusingAIandAutonomousLabs.(NationalInstituteofStandardsandTechnology,

Gaithersburg,MD),NISTSP1320.

/10.6028/NIST.SP.1320

AuthorORCIDiDs

HowieJoress:0000-0002-6552-2972

ZacharyTrautt:0000-0001-5929-0354

AustinMcDannald:0000-0002-3767-926X

BrianDeCost:0000-0002-3459-5888

AaronGiladKusne:0000-0001-8904-2087

FrancescaTavazza:0000-0002-5602-180X

ContactInformation

howie.joress@,francesca.tavazza@

i

NISTSP1320August2024

Abstract

WiththegoalofadvancingUScompetitivenessandexcellenceinthematerialsandmanufac-turingindustries,wepresentourvisionfortheNationalCenterforAutonomousMaterialsScience.Theobjectiveofthiscenteristoenableandpromotetheuseofautonomousmethodologiesformaterialsscienceinindustrialapplications.Virtuallyeveryindustryisdefinedandlimitedbythematerialsandprocessingavailabletoit.Thefoundationofmaterialscienceistoelucidateandexploittherelationshipsbetweenstructure,processing,andpropertieswithinmaterialswhichaffectitsultimateperformance.Usingtraditionalresearchapproachesthisistypicallyalengthy,costly,andcomplexprocess.Newparadigmsofconductingresearch,leveragingcuttingedgeAIandautomationtechnology,arebeingdevelopedthatcanaddressgrandchallengesinmaterialsresearchanddevelopmentthatareotherwiseintractable.Thesenewresearchparadigmsadditionallyacceleratetheacqui-sitionofknowledgeandunderstandingofcriticalmaterialsproblems.Unfortunately,mostindustrialR&Daswellassomeacademicresearchdoesnottakefulladvantageofthesenewresearchparadigms.TheU.S.needstofostertheadoptionofthesenewresearchparadigmsasanationaleffortinordertoaccelerateandmaintainglobaltechnologicalleadership.OurvisionisfortheUStosupportaninitiativethatwouldfacilitatetheindustry-wideadoptionofthesenewmaterialsscienceresearchparadigms.

Keywords

Autonomousmaterialsscience;Self-drivinglab;Testbedfacility.

NISTSP1320August2024

ii

TableofContents

ExecutiveSummary

1

1.Introduction

1

1.1.FillingtheMaterialsTechnologyGap

1

1.2.AutonomousR&D:ANewParadigm

1

1.3.BenefitsofAutonomousMethods

2

1.4.AutonomousMethodscanSolveGrandChallengesinMaterialsandManufac-

turing

3

1.5.CurrentChallengestoAutonomousImplementation

4

1.5.1.Lackofstandardizedprotocols

4

1.5.2.Advanceddomainspecificalgorithms

5

1.6.ThePathForward

6

2.MajorGoals

7

2.1.RevolutionizingHowU.S.IndustryDoesMaterialsResearchandDevelopment

7

2.2.EliminatingtheFormidableBarrierstoAutonomousTechnologyAdoption

7

2.3.MakingPreviouslyImpossibleMaterials“GrandChallenges”Tractableand

Achievable

7

3.AVisionforaNationalCenterforAutonomousMaterialsScience

8

Approach1:BuildNewWorld-ClassAutonomousTestbedFacilities

8

Approach1.1:DevelopmentSandbox

8

Approach1.2:ExemplarSystem

8

Approach1.3:UserFacility

9

Approach1.4:DataGenerator

9

Approach2:DevelopaStandards-BasedModularLaboratoryEcosystem

9

Approach2.1:DevelopSampleManagementStandards

10

Approach2.2:DevelopInstrumentControlandCommunicationStandards

10

Approach2.3:DevelopDataandKnowledgeManagementStandards

11

Approach2.4DevelopAlgorithmandModelIntegrationStandards

11

Approach3:EmpowerPublic-PrivateCooperationviaNewConsortia

12

Approach4:IncentivizeEcosystemDevelopmentandHigh-RiskActivitieswith

DirectedFundingOpportunities

12

Approach5:NextGenerationWorkforceEducation

13

iii

Approach6:End-Usersupport

13

4.Impact

13

5.WholeofGovernmentApproach

14

6.Resources

14

6.1.BuildingAcquisition,Construction,andOutfitting:$200M

15

6.2.ExperimentalHardwareAcquisition:$120M

15

6.3.ComputationalHardwareAcquisition:$100M

15

6.4.Personnel:$180M($18Mperyear)

15

6.5.IncentivesforIndustryStakeholders:$200M

16

6.6.Otheroperatingcosts:$200M($20Mperyear)

16

7.Timeline

16

References

17

ListofFigures

Fig.1.Themajorfunctionalcomponentsofanautonomouslaboratory.Adoptedfrom

Ref[1]

3

iv

NISTSP1320August2024

Acknowledgments

TheauthorsthankDHolbrookandJWarrenforusefuldiscussionandfeedback.

1

NISTSP1320August2024

ExecutiveSummary

InresponsetotheSenateRoadmapforArtificialIntelligencePolicyintheU.S.,wepresentavisiontoempowerandbuildtheproposed”testbedtoidentify,test,andsynthesizenewmaterialstosupportadvancedmanufacturingthroughtheuseofAI,autonomouslaboratories,andAIintegrationwithotheremergingtechnologies,suchasquantumcomputingandrobotics.”

MaterialsTechnologyGap:Allmoderntechnologyislimitedbythecapabilitiesofcurrent-generationmaterials.Currently,U.S.industriesrelyonmethodsspanningempiricaltrialanderrortophysics-anddata-basedmethodstodevelopnewadvancedproducts.Unfortunately,somelegacyR&DapproachesmaynotbecompetitiveinaworldcurrentlyundergoinganAIrevolution.ThiseffortaimstoenableU.S.industriestoleveragethemostpowerfulparadigmsofR&Dapproachestodevelopnewadvancedandsustainablematerialsandproducts.ThesenewapproachesexpandthelimitsofautonomouslaboratoriesandscientificallyexplainableAItosolveournation’smosturgenttechnicalchallenges.

Opportunity:Autonomousexperimentation(AE),orself-drivinglaboratories,combinesAIandautomationwhileleveraginghumanintuitionandcreativitytoguidecampaignsofexperiments.AEcanreducethetimeandresourcesneededtodiscovernewmaterialsbyordersofmagnitudeandreducethetime-to-marketforcriticaltechnologiesbasedonthem.Further,AEhasthepowertoaddressproblemsthatareotherwiseintractablycomplexorevermutating.MaterialsR&Dinvolvescombiningandprocessingdifferentkindsofrawmaterialsindifferentways,andtherearemorepossiblecombinationsthanstarsintheknownuniverse.Thisnearinfinitesolutionspaceistheunderlyingbasisofallgrandchallengesinmaterialsandmanufacturing.Aninitiativeisneededtodeveloptoolsthatenableindustrytoautonomouslynavigatethisinfinitesolutionspaceefficientlywithautomatedlaboratories.AIhasthepowertodiscerncomplexpatternsinlargedatasetsincludingmultiplestreamsofinputmaterialsdata.Automationenablesvitalmaterialsdatatobecollectedrapidlyanddynamically.ToenableUSIndustrytomaintaintechnicalleadershipindevelopingcutting-edgematerialsandtechnologies,theUSmustbecomealeaderinAE.

AutonomousExperimentationChallenges:WhiletheunderlyingtechnologyforAIandautomationexistsinmanydomains,theirapplicationtomaterialsresearchisstillinitsinfancy.Currently,thereisnostandardizedAEecosystemformaterialsR&D,whichisamajorobstacletoindustrialadoption.Creatingastandardsbasedecosystemwilldramaticallyreducethecostofengineeringaplatform,aswellasreducetheriskofobsolescencebyensuringitisexpandableandupgradeable.Relatedly,adefinedstandardwillempowerequipmentandsoftwarevendorstodesigntheirproductsforautonomousintegration,reducingtheircostandliabilityassociatedwithbespokeengineeringandensuringacustomerbasefortheirproducts.Beyondthesestandardizationchallenges,therearemanybasicandappliedresearchchallengestoincreaseefficacyofAE,includingdevelopmentofnewexplainableAIalgorithmsaswellassynthesismethodologyandmaterialsmetrology.

Goals:PromoteUStechnologicalleadershipby:

1.RevolutionizinghowU.S.industrydoesmaterialsresearchanddevelopment:ThiseffortseekstopropelU.S.industriesintomorepowerfulparadigmsofresearch,acceleratingthediscoveryandcommercializationofnewtechnologiesanddevicesthatsolvepressingsocietalneeds.

2

NISTSP1320August2024

2.Eliminatingtheformidablebarrierstoautonomoustechnologyadoption:Acriticaloutcomeofthiseffortwillbetheavailabilityofoff-the-shelf,autonomoussolutionsthatwilldrasticallyreducetheinvestmentandriskassociatedwithimplementingautonomousR&Dworkflows,whichiscriticalforsmall-andmedium-sizedbusinesses.

3.Makingpreviouslyimpossiblematerials“GrandChallenges”tractableandachievable:Thiseffortseekstodeveloptoolstonavigatenearinfinitesolutionspacesinarapid,intelligent,andmeaningfulway.

Approach:

1.Buildanewworld-classcentertohostautonomoustestbedfacilities:WeenvisiontheconstructionofanewNIST-lednationalcentertoworkside-by-sidewithindustry,government,academiatodevelopastandards-based,modularecosystemofnewhardware,software,andmethodsforAEinmaterialsR&D.Thiscenterwillhostatestbedthatwillserveasatechnologydemonstrator,technologyincubator,nationaluserfacility,andreferencedatagenerationplatform.Thistestbedwillconsistofasetofautonomous-readyscientificinstrumentsforon-demandmaterialssynthesisandmaterialscharacterization.Theecosystemwillbedesignedtobereadilyimplementableinlabsacrosstheresearchsectorincludingwithinindustry,government,andacademia.Thistestbedandecosystemwillempowerhuman-AIteamingformaterialsresearch.

2.Developamodularautonomouslaboratoryecosystem:Justastheinternetrevolutionwastheoutcomeoflow-levelcommunicationstandards,weaimtoinitiatealaboratoryrevolutionthatwillbepoweredbyastandards-basedmodularlaboratoryecosystem.Inpartnershipwithindustrystakeholders,NISTwillleaddevelopmentofthisnext-generationlaboratoryecosystemthatisbasedonandenabledbyvoluntaryconsensusstandards.Thesestandardswillenablecomponentsfortheecosystem,producedbyindustrypartners,tobemodularandplug-and-playandwillreducetheriskforproductvendors.NISTandpartnerswillusethetestbedasanenablingtooltocoordinateanddevelopstandardsandprotocolsforindustry-wideinteroperability,includingsamplemanagement,instrumentcommunication,datamanagement,andAIalgorithms.Developmentofnewmarketableecosystemcomponentsbyindustrywillbeincentivizedbydirectfinancialsupportthroughcompetitivedevelopmentgrantsandothermechanisms.Thiseffortseekstoinitiatearevolutioninhowlaboratoryinfrastructureisdeveloped,commercialized,andused.

3.Cultivatepublic-privatepartnershipsandworkforcedevelopment:WeenvisionNISTleadingthecoordinationofmulti-agencyeffortsandpublic-privatepartnershipstorealizethefullpotentialoftheAEecosystemwithinU.S.industry.ThisNIST-ledcenterwouldworktoprovideorganizationalframeworksandincentivemechanismsforprioritycollaborations.Thesecollaborationswouldrevolvearoundthedevelopmentofnewhardware,software,andalgorithmstoexpandtheecosystemofAEcomponents.Educatingthenext-generationworkforcewillbecriticaltoensurethefullandsustainableimpactofthisAErevolution.Thetestbedwillalsoserveasaplatformforhands-onpracticaltrainingacrossarangeofeducationallevels,fromvocationaltodoctoralprograms.

TimeandResources:Weestimatethatthisvisionwilltake10yearstocometofullfruitionandrequire$1Bforthecenter.

1

NISTSP1320August2024

1.Introduction

1.1.FillingtheMaterialsTechnologyGap

AsexpressedbestbyGreenetal.

[2]:

Materialsaretechnologyenablers.Therewouldbenoskyscraperswithoutsteelgirders.Therewouldbenocommercialaviationindustrywithouthigh-strengthaluminumalloysandadvancedcomposites.Therewouldbenoinformationagewithoutsilicon.Therewouldbenomobilephoneswithoutfunctionalceramics.Therewouldbenosolarelectricitywithoutphotovoltaicmaterials.Therewouldbenomodernmedicinewithoutbiocompatiblesoftmaterials.Virtuallyeveryindustryisdefinedandlimitedbythematerialsandprocessingavailabletoit.

Thefoundationofmaterialscienceistoelucidatetherelationshipsbetweenstructure,processing,andpropertieswithinmaterialsandhowtheyaffecttheirultimateperformance.Theworkofmaterialsdesignistoharnesstheknowledgeofthoserelationshipstomakematerialswiththeneededpropertiesandperformancetomeetcriticalsocietalneeds.

Thespaceofpossiblematerialprocessing,theresultantstructures,andtheirrespectivepropertiesistoolargetoexploreandoptimizeefficientlywithtraditionalresearchmethods.Mosttechnologicallyrelevantmaterialsaretoohierarchicallyintricateforeffectivetheory-drivenapproaches,andbrute-forcetrial-and-errorapproachessufferfromthe“curseofdimensionality”,theexponentialexplosionofpossibilitieswithincreasingnumberofdesignparameters.Yet,thesearestillthedefaultapproachestomaterialsdiscoveryandprocessingoptimizationproblemsinmanyindustries.Autonomouslaboratoriesimplementhardwareandsoftwaretoolsthathelpnavigatethesearchspaceofthesematerialssciencechallengeswithmoreefficacy.Theoverarchingobjectiveoftheeffortdescribedhereistodemonstrateandfacilitatetheuseofautonomouslabsformaterialsscienceinindustrialapplications.Ourvisionistocreateanationalcentertoenableandpromoteautonomousmethodsformaterialsscience.

1.2.AutonomousR&D:ANewParadigm

Tomeetthematerialsneedsofthe21stcenturyitisnecessarytochangethewaymaterials

R&Disdone.Researchistypicallycategorizedinto4majorparadigmsofscientificstudy:(i)empiricalscience,(ii)theorydrivenscience,(iii)computationalscience(e.g.,simula-tions),and(iv)datadrivenscience.U.S.industriesareatvariouslevelsofembracingeachparadigm.ThefederalgovernmenthasbeenactivelyincentivizingadvancementofthematerialssciencecommunityforsometimethroughprogramssuchasfundingIntegratedComputationalMaterialsEngineering(ICME)

[3]

effortsandtheMaterialsGenomeInitiative(MGI)

[4]

.Theseinitiativeshavepropelledindustryforwardinharnessingcomputationalanddatadrivenmethods.The4thparadigmleverages”bigdata”andmachinelearningto

2

NISTSP1320August2024

makepredictionsandinterpolationsfromwhatisalreadyknown,oftenincasesthataretoocomplextocreatetheorybasedmodels.

[5]

However,manymaterialsproblemsrequiretraversingaparameterspacethathasnotbeentraversedbefore,whereevenconventional(4thparadigm)“bigdata”extrapolationswillfail.Furthermore,manyimportantmaterialsproblems(parameterspaces)areimpossibletoexplorewithanyofthe4existingparadigmsofscientificdiscovery.

Currently,mostR&Dexperimentsarecarriedoutusingindependentinstrumentswithalargeamountofmanualinteractionforoperationssuchassamplesynthesis,characterization,dataanalysis,andknowledgeextraction(e.g.,findingpatternsandtrends).Autonomousexperimentation(AE),colloquiallyreferredtoasself-drivinglabs,offersanewoperationalparadigmthatsystematicallygeneratesthemostinformativedatatowardtheendgoaloftheparticularproblemathand,greatlyincreasingtheefficiencyofresearchbudgets.AEcanleveragecutting-edgeadvancesinautomatedexperimentationandmachinelearn-ingalgorithms,alongwiththeoreticalcomputation,forrapidlyacquiringnewmaterialsknowledge.Thisfreeshumanscientiststofocusonimbuingthesystemwithintuitionand

leveragingtheircreativitytoguidethesystemtowardsimportantwork.Fig.

1

illustrates

thisconceptually.Thecoreofanautonomoussystemisatightlycoupledfeedbackloop.Tostarttheloopanexperimenterframestheproblemwithasetofobjectivesandcon-straints.Priorknowledgeincludingdatabasesofknownmaterialsandtheirproperties,physicochemicalheuristics,andphysicallawscanalsobeprovidedtotheplatform.TheAI-basedagentwillthentakeallofthisinformationanduseittogenerateacomputationallyinexpensiverepresentativemodelofthesystem,thenuseittomakepredictionsacrossthevastparameterspace.Basedontheobjectiveandthemodel,theagentwilldecidewhatnewdatawillprovidethemostcriticalpieceofinformationtoincreaseitsknowledgeandreachitsobjective,typicallythroughsomecombinationofexplorationandoptimization.Theautomatedsystemwillthenperformanexperimentortheorybasedsimulationtoproducethatknowledge.Theplatformthenanalyzes,interprets,andstoresthatdata.TheAIagentwillthenlookattheupdateinformationandmakeadecisiononthenextdatapoint.Thisloopisrepeateduntilsufficientinformationisgatheredtoachievetheobjective.

1.3.BenefitsofAutonomousMethods

Autonomousmethodshaveseveraladvantagesovertraditionalresearchmethods.Theuseofactivelearning,alongwithinclusionoftheory-basedmodeling,cangreatlyincreasethevalueofeachexperimentaldatapointcollected.Conversely,thesemethodsreducetheamountoflow-valuedatathatneedstobecollectedinordertogeneratethesameknowledge.Particularlyinfieldslikematerialsscience,generatingdatacanbefiscallyexpensiveandresourceintensive.Reducingthenumberofexperimentsalsoimprovestheenvironmentalsustainabilityofresearchbyreducingtherequiredexperimentalpre-cursorsandexperimentalwaste.Furthermore,AI-baseddecision-makingsystemswithwell-characterizedassumptionscanreducetheunquantifiableeffectsofbiasonthepartofhumanexperimenters,bothondecisionmakingandanalysis.

3

NISTSP1320

August2024

Fig.1.Themajorfunctionalcomponentsofanautonomouslaboratory.AdoptedfromRef

[1]

Collectingdataatscalessufficienttomeetsocietalneedsrequiresagileexperimentalcapabilitiestodynamicallyexploredisparatehypotheses.Robotic,automatedplatformsachievethisbyquicklyandaccuratelyexecutingexperimentsondemand.Theseautomatedplatformsbringseveraldistinctadvantages:First,thedataisgeneratedwithroboticcontrol,whichhastheadvantageofbeingmoreprecise,traceable,repeatable,andsystematic.Second,thedigitallynativedataproducedbyautomatedlabsstreamlinescollectionandrecordingofcrucialmetadataabouttheexperiment.Third,theselaboratoriescanalsorunwithouttheneedfordowntimerequiredbyhumanlabworkers.Thetimeandcostsavingsassociatedwiththeseplatformsisnotsimplyadifferenceindegree,butinmanycasesisadifferenceinkind;makingproblemswhichwerepreviouslyintractablycomplextractable.

1.4.AutonomousMethodscanSolveGrandChallengesinMaterialsandManufacturing

MaterialsR&Dinvolvescombiningandprocessingdifferentkindsofrawmaterialsindifferentways,andtherearemorepossiblecombinationsthanstarsintheknownuniverse.Thisnearinfinitesolutionspaceistheunderlyingbasisofallgrandchallengesinmaterialsandmanufacturing.TherearesometechnicalareaswhereAEplatformsmaybetheonlywaytomakematerialsdesigntractable.Someareassuchashighentropyalloys—alloyswithmultipleelementsinlargeproportionswhichpromiseextraordinarypropertiesforapplicationsasvariedashypersoniccoatingsandwatersplittingcatalysts—existinsuchalargecompositionalandprocessingdesignspacethatcollectinginformativedataonthesealloysthroughstandardexperimentalapproachesisintractable.Anotherplaceautonomous

4

NISTSP1320August2024

platformsarecrucialisindomainswheretheproblemisconstantlymutating.Twoexamplesofthisaremetals-basedadditivemanufacturing(AM)andcirculareconomy.Intheformerexample,optimizationofAMprocessingparametersisacomplexproblem.Eachuniquecombinationofpartgeometry,materials,andprintingplatformrequiresanewoptimization.Efficientlymappingthisoptimizationsurfaceistheonlypathforbeingabletodesignpartswithoutdenovoexperimentationforeachpartandprocess.Similarly,recyclingofmaterialsaspartofacirculareconomyrequirescontinuallyoptimizingtheprocessingtoaccountfortheevermutatinginputmaterialstream,whichcanhavestrongimpactsontheend-productpropertiesandperformance.Understandingwhateffectsarecausedbythechanginginputstream,whatmaterialscanbemade,andwhatadjustmentstotheprocessingareneededisanongoingdesignchallenge.Inallofthesecases,thematerialsphysicsistoocomplextobemodeledusingcurrentlyavailable,first-principlesbasedapproaches,soexperimentationisanecessarydrivingforceforinnovation.

1.5.CurrentChallengestoAutonomousImplementation

1.5.1.Lackofstandardizedprotocols

R&DlabsthatwouldliketotakeadvantageofthebenefitsthatAEcanprovidefaceavarietyofchallenges,andtheirassociatedcosts,whenitcomestoimplementinganautonomousplatform.Manyofthesechallengeshavebeenhighlightedinarecentgovernmentwidereport:Ref.

[6]

.Therearecurrentlynoagreedprotocolsforhowsamplesshouldbetransferredbetweeninstruments,norstandardizedprotocolsforinteroperabilityfordatasharingorinstrumentcontrol.Thisleavesresearcherswithtwooptions:seekanoutsideequipmentvendortoputtogetheraninstrument,typicallyatgreatcost,ortakethetimetocobbletogetherinstrumentsin-house.Becausethereisnostandardization,theseplatformsaretypicallydesignedforasinglepurpose,meaningtheengineeringworkputintothemcannottypicallybeleveragedforfutureplatforms.Furthertheplatformsaretypicallybuiltforonepurposeandwhenthescopeofinquiryshiftstheplatformbecomesobsolete,necessitatingamajorredesign.

Thecurrentrarityofthesesystemshasledtoandiscausedbya“chickenandeggproblem”betweenvendorsandend-users.Becauseofthelargeinvestmentcurrentlyneededtoprocureordevelopanautonomoussystem,currentlyfewend-usersareattemptingtouseoneaspartoftheirR&Defforts.Asaresult,vendorsdonotbelievethereisamarkettocreatehardwareandsoftwaretosupportthesetypesofsystemsandthereforearereluctanttomaketheinvestmenttodeveloptheseproducts.Inturn,thismakesbuildinganautonomoussystemmorechallengingandexpensive.

AclearexampleofthiswasdiscussedindetailattheOctober2023AutonomousMethodolo-giesforAcceleratingX-rayMeasurementsworkshophostedbyNISTandtheInternationalCenterforDiffractionData.Severalmanufacturersofcommonx-raydiffractometerswerepresentandmanydescribedtheirreluctancetocreateinterfacesnecessarytoenableau-tonomouscontroloftheirinstruments.Theclearmessagewasthat,atpresent,therewas

5

NISTSP1320August2024

notsignificantcustomerrequestforthesefeatures(butnotnone).Thefurtherconcernwasthatthelackofastandardinterfacemeantdifferentcustomerswouldrequestdifferentinterfaces,requiringrepetitivecustomengineering.Arelatedriskthatwasdiscussedistheliabilityforvendorsthatchangestosoftwarethatinteractswiththeirinterfacesbutisoutoftheircontrolwouldinverselyaffectthefunctionalityoftheirtool;aproblemforwhichcustomerswoulddemandfixes.Giventhesechallenges,theseequipmentmanufacturersdecidedthat,forthemoment,theywouldavoidofferingthisfeatureset.

Anotherchallengethatisslowingthewidespreadadoptionofthesetechnologiesisthetechnologytransferoftools,bothhardwareaswellassoftware,fromthemanyacademicresearchlabsworkingintheAEspacetoindustry.Thisisaparticularissueforsoftwaretools;manytoolswrittenbystudentsareoftennotmaintainedataprofessionallevelnecessaryforadoptionbyequipmentmanufacturersandindustrialusers.

1.5.2.Advanceddomainspecificalgorithms

Autonomousphysicalscienceisstillinitsearlystages.Inadditiontotheengineeringchal-lengesdescribedabove,thereareseveralopportunitiesforfurtherinnovationtoimprovetheseautonomousworkflows.Someoftheseinnovationscanboostthescopeandspeedofwhatcanbeaccomplishedwithinth