2026年物理人工智能报告(英文版)_第1页
2026年物理人工智能报告(英文版)_第2页
2026年物理人工智能报告(英文版)_第3页
2026年物理人工智能报告(英文版)_第4页
2026年物理人工智能报告(英文版)_第5页
已阅读5页,还剩134页未读 继续免费阅读

付费下载

下载本文档

版权说明:本文档由用户提供并上传,收益归属内容提供方,若内容存在侵权,请进行举报或认领

文档简介

PhysicalAI

Takinghuman-robotcollaborationtothenextlevel

2

PhysicalAI:Takinghuman-robotcollaborationtothenextlevel

Tableofcontents

04

Whoshouldread

thisreportandwhy

36

PhysicalAIisagame-changerforindustry

56

ThegrowingimperativetoadoptphysicalAI

08

24

WhyphysicalAIis

ataninflectionpoint

Executivesummary

CapgeminiResearchInstitute2026

3

CapgeminiResearchInstitute2026

PhysicalAI:Takinghuman-robotcollaborationtothenextlevel

104

Conclusion

70

96

105

ReseaΓchmethodology

ScalingphysicalAIgoes

beyondtechnology,spanningsafety,cybersecurity,

Recommendations:

Acceleratingthe

physicalAIrevolution

78

Humanoidssetthestage

forgeneral-purposerobotics

regulation,andoperations

CapgeminiResearchInstitute2026

4

PhysicalAI:Takinghuman-robotcollaborationtothenextlevel

WhoshouldΓeadthisΓepoΓtandwhy?

Thisreportisintendedforseniorexecutives

shapingtheirorganizations’approachto

roboticsandautomation.Itexamineshow

physicalAIistransformingrobotics–fromthecapabilitiesitenablestothevalueitunlocks,

thetimelinesforadoption,andthebarriersthatmustbeaddressedtoscaledeploymentssafelyandeffectively.Itwillbeparticularlyrelevanttotechnologyandinnovationleaders(including

chieftechnologyofficers,chiefinnovation

officers,chiefdigitalofficers,andheadsofAIorrobotics),aswellasmanufacturing,supply

chain,andlogisticsleadersresponsibleforroboticsstrategyanddeployment.

Asroboticsexpandsintoconsumer-facingandserviceenvironments–suchashealthcare,

retail,hospitality,andentertainment–the

reportisalsorelevanttochiefproductofficers,productstrategists,andexperiencedesign

leaderswhoareshapinginteractionsbetweenpeopleandintelligentmachines.

Inaddition,thereportprovidespractical

guidanceforCROsandsafetyorregulatory

leaderspreparingtheirorganizations

forwiderroboticsadoption–including

implicationsforgovernanceandriskoversight.

Thisreportdrawsonaglobalsurveyof

1,678seniorexecutivesacross15industries,complementedbyin-depthinterviewswithindustryexperts,robotmanufacturers,

foundation-modelstartups,technologyproviders,investors,andacademics.

Pleaseseetheresearchmethodologyattheendofthereportformoredetails.

CapgeminiResearchInstitute2026

5

PhysicalAI:Takinghuman-robotcollaborationtothenextlevel

WeextendouΓsinceΓethankstothemanyexpeΓtsfΓomindustΓyandacademiawhoshaΓedtheiΓinsightswithus

DanielaRus

Director,ComputerScienceandArtificialIntelligence

Laboratory(CSAIL),MIT

RebeccaYeung

StrategicAdvisoratDexterityandformerCorporateVicePresidentforOperationsScienceand

AdvancedTechnologyatFedEx

DeepuTalla

VPandGM–Robotics&EdgeAI,NVIDIA

AshutoshSaxena

FounderandCEO,

TorqueAGI

SanjayAggarwal

VenturePartner,

F-PrimeCapital

6

PhysicalAI:Takinghuman-robotcollaborationtothenextlevel

AngeloCangelosi

Co-DirectoroftheManchesterCentreforRoboticsandAI,

UniversityofManchester

NageshPuppala

GeneralManager,RoboticsandPhysicalAI,ClientComputing

Group,IntelCorporation

MiladMalekzadeh

Co-FounderandVicePresidentAI,

NeuraRobotics

PedroZheng

SeniorRegionalSalesManager,

UnitreeRobotics

AntoPatrex

FounderandCEO,

CosmicBrainAI

JimMa

RegionalTechnicalDirector,

UnitreeRobotics

DanielJacker

CEO,ZaiNar

JulienPerrin

COO,Niryo

DirkGeiger

SeniorDirectorandTeamLead–HumanoidRobotics,

InfineonTechnologies

VikiYang

OverseasSalesDirector,

UBTECHRobotics

CapgeminiResearchInstitute2026

7

PhysicalAI:Takinghuman-robotcollaborationtothenextlevel

Executivesummary

"ThelastdecadeofAIwas

aboutinformation.Thecomingdecadewillbeaboutaction."

PhysicalAItakesAIbeyondscreensintotherealworld–enablingmachinestoperceive,reason,andactautonomously.Thisreportfocuseson

itsapplicationinrobotics,wherephysicalAI

representsafundamentalshift:fromrobots

thatfollowfixed,pre‑programmedpathsto

robotsthatcangeneralizeacrosstasks,perceiveandnavigatecomplexenvironments,make

context-awaredecisions,andadapttoreal-worldvariation.Thisenablesrobotstofunctionin

farmorediverseanddynamicenvironments,

RebeccaYeung

StrategicAdvisoratDexterityandformerCorporateVicePresidentforOperationsScienceandAdvancedTechnologyatFedEx

expandingtheirreachacrossnearlyeverymajorindustryandunlockingsolutionstoproblems

earlierautomationcouldn’taddress.

8

PhysicalAI:Takinghuman-robotcollaborationtothenextlevel

Executivesummary

TraditionalroboticsversusroboticspoweredbyphysicalAI:AcompaΓison

Traditionalrobotics

Keyfeatures

PhysicalAI-poweredrobotics

Limitedperception–senseswithoutinterpretation

Perception

Perceivestheenvironmentthroughrich,multi-modalsensing

(vision,depth,touch,audio)andinterpretscomplexenvironments

Worksonlyinstructured*environments(consistent,predictablesettings)

Adaptability

Operatesinunstructured**environments(messy,variable,dynamicsettings),includingpreviouslyunseensituations

Hasnorealautonomy;followspre-programmedinstructions

Autonomy

Makescontext-awaredecisionsinrealtime

Noongoinglearning;behaviorisstaticunlessreprogrammed

Learningcapability

Learnsfromdemonstrations,simulations,andexperience,

improvingperformanceovertimewithoutmanualreprogramming

Designedforasingle,specializedtask

Generalization

Handlesmultiplescenariosonasinglerobot;generalizeslearnedskillstonewtasksandunfamiliarsituations

Robotsoperateindependentlywithnoknowledgesharing

Collectivelearning

Robotsshareskillsa∩dlear∩i∩9sacrossaHeet

Requiresprecise,codedcommands

Natural-languageunderstanding

Understandsnaturallanguageinstructions

Canexecuteassemblyonlyinastrictlyprogrammedmanner;failseasilyifpresentedwithanyslightdeviationfromprogrammedsequence

Example

Capableofadaptingautonomouslytovariationinassemblyprocessandsupportstailoredassemblybyadjustingdynamicallytoeachuniqueproduct

*Structuredenvironments:Environmentswherethelayout,tasks,andconditionsarepredictableandconsistent,allowingrobotstofollowfixedpathsandroutineswithlittlevariation.Examples:assemblylines,controlledwarehouseaisles.

**Unstructuredenvironments:Environmentsthatarevariableandunpredictable,whererobotsmustadapttochangeanduncertainty.Examples:retailfloors,hospitals,farms,constructionsites.

ForamoredetaileddescriptionofphysicalAI,itsapplicationinrobotics,andindicativeindustryusecases,pleaserefertotheAppendix.

Source:CapgeminiResearchInstituteanalysis.

9

PhysicalAI:Takinghuman-robotcollaborationtothenextlevel

Executivesummary

TounderstandtheimpactofphysicalAIon

roboticsandthevalueitcanpotentiallyunlock,thisreportdrawsonaglobalsurveyof1,678executivesacross15industries,complementedbyin-depth

interviewswithexpertsacrossthephysicalAI

androboticsecosystem(pleaseseetheresearchmethodologyformoredetails).

PhysicalAIisataninflectionpoint

MultimodalfoundationmodelsaΓeΓedefining

robotintelligencebyenablinggeneralization

acrosstasksandenvironments.Theseadvancesareallowingrobotstoadapttounfamiliarsituations

withouttaskspecificΓepΓogΓamming,extendingdeploymentintounstructuredenvironments

–messy,dynamicsettingsthatearlierrobotic

systemscouldnothandle.Inparallel,advancesinsimulationareshorteningrobottrainingcycles,whileanAl-Γobot-dataflywheelisacceleΓating

improvementwitheveryreal-worlddeployment.Combinedwithfallingcostsofkeyhardware

componentssuchassensors,actuators,andelectricmotors,andcommercialmodelssuchasrobotics-as-a-service(RaaS),theseshifts

areloweringbarrierstoadoption.Atthesametime,demographicandeconomicpressures

–includingagingworkforcesandpersistent

laborshortages–areintensifyingdemandfor

roboticsystemscapableoftakingonrolesthataΓeincΓeasinglyhaΓdtostaff.RecoΓdventuΓe

capitalinvestmentintophysicalAIandroboticsisaddingtothemomentumbehindtheseshifts.

Agame-changeracrossmultipledimensions

PhysicalAImarksastepchangefromearlier

automation.Byenablingrobotstointerpret

context,adaptinrealtime,andoperatein

unstructuredenvironments,physicalAIpromotesthemfrompassivetoolstoactivecollaboratorsintheworkspace–openingthedoortoareimaginedworkenvironment,inwhichhumans,robots,

andAIagentsworkintandem.Atthesametime,

physicalAIallowsroboticstoscaleasashared

intelligenceplatform,withlearningandcapabilitiescompoundingacrossdeployments.Indoingso,

physicalAIextendstheagenticparadigmintotherealworld,enablingrobotstoactasembodied

AIagentscapableofplanning,orchestrating,and

CapgeminiResearchInstitute2026

10

PhysicalAI:Takinghuman-robotcollaborationtothenextlevel

Executivesummary

executingcomplexphysicaltasks.Overtwo-thirds(67%)ofexecutivesviewitasgame-changingfortheirindustryandmostbelieveitwillbecomea

criticaldriverofcompetitiveness.

67%

ofexecutivesviewphysicalAIasgame-changingfortheirindustry

PhysicalAI’svalueismulti-faceted.Executivesexpectthestrongestgainsinproductivity,

e代cie∩cy,a∩dquality,alo∩9side9reater

operatio∩alresilie∩cea∩dHexibilityasadaptiverobotshelporganizationsmanagevolatilityandreco∩fi9ureoperatio∩squickly.PhysicalAIalso

improvesworkplacesafetyandreducesphysicalstrain,asrobotsincreasinglytakeonhazardousandphysicallydemandingtasks.Beyond

operationalimpact,physicalAIisopeningnew

growthavenues:nearlyfourintenexecutives

i∩spectio∩,alo∩9sidesector‑specificapplicatio∩ssuchasdynamicassemblyinmanufacturing,

healthcareandeldercareinthepublicsector,anddisaster-damageassessmentininsurance.

expectnewrevenueopportunities,and60%

believeitwillenableroboticsinareasthatwerepreviouslyimpossibleorimpractical.High-impactusecasesspanhazardousoperations,

micro‑lo9istics,pick‑a∩d‑place,a∩dfield

64%

ofexecutivesbelievephysicalAIwill

becomeacriticaldriverofcompetitiveness

CapgeminiResearchInstitute2026

11

PhysicalAI:Takinghuman-robotcollaborationtothenextlevel

Executivesummary

ThereisagrowingimperativetoadoptphysicalAI

PhysicalAIadoptioniswellunderway:nearly

eightintenorganizations(79%)arealready

engaging,with27%deployingorscaling,and65%expectingtoreachscalewithinfiveyears.The

primarycatalystsarestructural:laborshortages

(74%)andrisinglaborcosts(69%).Inthenear-

term,growthwillcomefromfamiliar,proven

formfactorsfortask‑specificapplications.As

foundationmodelsmatureandadoptiondeepensacrossindustries,entirelynewcategoriesofrobotsarelikelytoemerge–purpose-builtforvaried

environments,complextasks,andnewmodesofhumancollaboration.Humanoids,despite

substantialinvestment,remainalonger-termbet,askeychallenges–includingtechnicalmaturity

(reliabilityanddexterity),safety,andcost-to-ROIviability–muststillbeaddressed.

Near-termgrowthwillcome

fromestablishedformfactorsfortask-specificapplications;asfoundationmodelsmature,newpurpose-builtrobot

categoriesarelikelytoemerge

ScalingphysicalAIgoesbeyond

technology–italsorequiresbuildingsafety,cybersecurity,regulatory,

andoperationalreadiness

Inpractice,scalingphysicalAIdemandsmore

thanbetteralgorithms–itrequiresrethinking

howsystemsareengineered,secured,governed,andrun.Today’ssystemsdonotyetmeetthe

highreliabilitythresholdsofindustrialandothersafety-criticalsettings,anddexterityremains

limited.Progressisfurtherslowedbydatascarcity–real-worldphysicalinteractiondataisscarceandcostlytoobtain.Tokeeppeopleandassetssafe

whilecapabilitymatures,safetymustbeenforcedthroughdeterministicmechanismsindependent

CapgeminiResearchInstitute2026

12

PhysicalAI:Takinghuman-robotcollaborationtothenextlevel

Executivesummary

oftheAIlayer.Further,asrobotautonomy

9rows,cybersecurityexposurewide∩s,requiri∩9controlsthatpreventunauthorizedaccessandmanipulation.Regulatoryframeworkslagthe

realitiesofautonomousphysicalaction,leavingu∩resolvedquestio∩saboutaccou∩tabilitya∩dacceptablerisk.Operationally,enterprisesmustpla∩forhardwareco∩strai∩ts,ma∩a9i∩9Heetsatscale,strengtheningdataandAIgovernance,andreskillingworkforces.

Humanoidrobotsinspirestrong

industryconviction–butscaled

deploymentremainsalong-termbet

Twointhreeexecutives(67%)believehumanoidswillultimatelytransformtheirindustry,citingtheirabilitytooperateinhuman-builtenvironments

andtheirpotentialasgeneral-purposesystems;53%arealreadyinvestingorplantoinvest.

However,theconditionsforscalearenotyetinplace.While78%expecttodeployhumanoidsatscaleeventually,averagetimelinesextendtosevenyears,andonly30%seethembecomingviablegeneral-purposeworkerswithinthreetofiveyears.Tech∩olo9yimmaturity,hi9hcosts,uncertainROI,andsafetyconcernsremain

si9∩ifica∩tbarriers,compou∩dedbyasocietal

readinessgap,with62%citingpublicacceptanceasacriticalhurdle.

67%

ofexecutivesbelievehumanoidswillultimatelytransformtheirindustry

CapgeminiResearchInstitute2026

13

PhysicalAI:Takinghuman-robotcollaborationtothenextlevel

Executivesummary

Recommendations:ActionstounlockthepotentialofphysicalAI

PhysicalAIadoptionisamulti-yearjourney,butthetechnologyismatureenoughtodelivertangible

valuetoday.

FivepΓioΓityactions:

1.Buildunderstanding:DevelopaclearviewofwhatphysicalAIenablestoday–itscapabilities,limits,anddata‑infrastructurerequirements.

2.Startwithconfidence-buildingusecases:

Beginwithfeasible,meaningfulapplicationsthatbuildfamiliarityandconfidence–suchasdull,dirty,ordangeroustasks.

3.Designthroughformexploration:Iteratewithmultipledesignconceptstoassesshowformshapestrust,interaction,andsuitabilityfordifferenttasksandenvironments,ratherthandefaultingtohumanoids.

4.Redesignworkflows:Reworkprocessesforhuman–robotcollaboration,with

clearhandovers,supervision,safety,andescalation.

5.Scaleviaplatforms:Createascalable

architectureforreusablerobotskills

andfleet‑levelorchestration,toenabledisciplinedscalingbeyondisolatedpilots.

Theseactionsmustbeanchoredintrust–throughclearsafety,governance,andhuman-oversight

guardrails–andsupportedbyongoingengagementwiththephysicalAIecosystemastechnologies,

standards,andregulationscontinuetoevolve.

CapgeminiResearchInstitute2026

14

Smartbet,onlyoption,orboth?BiopharmaR&DturnstoAI.

"PhysicalAImarksashiftfromsystemsthatdescribetheworldto

systemsthatcanactwithinit.Butweshouldstayclear-eyed.Roboticshasalonghistoryofoverpromising,whereearlybreakthroughscreatedexpectationsthetechnologycouldnotyetmeet.Whatisdifferent

todayisnotthehype,buttheconvergenceofAI,data,andengineeringmaturity.Theopportunityisreal,providedwefocusonwhatworksatscale,andgobeyondwhatlooksimpressiveindemos."

PascalBrier

GroupChiefInnovationOfficer,Capgemini

15

PhysicalAI:Takinghuman-robotcollaborationtothenextlevel

PhysicalAI–poweredroboticsinaction

PhysicalAIroboticsystemshavepotentialapplicationsacrosseverymajorindustry.Thefollowingexampleshighlighttheseapplicationsincomplex,dynamic,real-worldenvironments.

FiguΓe1.

ExamplesofphysicalAI–poweredroboticdeployments

IndustriesIllustrativecases

Warehousingandlogistics

Ultra,aUS-basedindustrialAIroboticscompany,haspartneredwithPhysicalIntelligence,aUS-basedstartupdevelopinggeneral-purposeroboticsfoundationmodels,todeployPl’sπ0.6modelonindustΓialΓobotsopeΓatinginlivewaΓehouseenviΓonments.ThemodelhasbeendeployedfoΓ

e-commeΓceoΓdeΓpacking,ataskthathashistoΓicallybeendifficulttoautomateduetolaΓgevaΓiabilityinitemtypes,defoΓmablepackagingmateΓials,andmulti-stepmanipulationthatcausesΓule-basedsystemstofail.Pl’sΓoboticfoundationmodelallowsUltΓa’sΓobotstopeΓceive,Γeason,andadaptinΓealtime.EaΓlydeploymentsshowUltΓa’sΓobotsachievinggainsinΓeal-woΓldautonomouspeΓfoΓmance,demonstΓatinghowphysicalAlcanunlock

waΓehousetaskspΓeviouslyconsideΓednon-automatable.1

FedExispartneringwithUS-basedroboticsstartupDexteritytopilot“superhumanoid”2robotsfortruckloading–oneofthemostcomplexand

physicallydemandingtasksinlogistics,aspaΓcelsvaΓywidelyinsize,shape,andweightandaΓΓiveinunpΓedictablesequences.TheΓobotsautonomouslyinteΓpΓettheincomingmixofpaΓcels,andstackthemintodense,stablewalls.UsingDexteΓity’sFoΓesightwoΓldmodel,theyevaluatehundΓedsof

possibleplacementsfoΓeachiteminmilliseconds,pΓedictinghoweachchoiceaffectstheintegΓityofthestack.ThisenablesΓapidhandlingofiΓΓegulaΓitems一wheΓetΓaditionalautomationstΓuggles一whileincΓeasingthΓoughputandΓeducingphysicalstΓaininhigh-volumeopeΓations.3

Manufacturing

FoxconnispartneringwithIntrinsic,anAlphabet-ownedcompanythatdevelopsAImodelsandsoftwareforrobotics,tohelprealizetheintelligentfactoΓyofthefutuΓe.ThecollaboΓationtaΓgetselectΓonicsassembly一afast-gΓowingsectoΓdΓivenbytheAlboombutstillconstΓainedbyΓigid

automationandmanualpΓocesses.ThepaΓtneΓshipaimstodeliveΓastepchangebyshiftingfΓompΓoduct-specificautomationthatΓequiΓesextensiveΓeengineeΓingacΓosspΓoductgeneΓationstomoΓegeneΓal-puΓposeintelligentΓobotics.lnitially,thecollaboΓationwilluselntΓinsic’sΓoboticsfoundationmodeltofocusoncΓiticalusecasesacΓossassembly,inspection,machinetending,andlogistics.4

Continuedonnextpage

CapgeminiResearchInstitute2026

CapgeminiResearchInstitute2026

16

PhysicalAI:Takinghuman-robotcollaborationtothenextlevel

Industries

Construction

Agriculture

TheconstΓuctionindustΓyfacesmountingpΓessuΓefΓomlaboΓshoΓtagesandincΓeasingdemandfoΓmoΓeefficientandsustainablebuildingmethods,whileincΓeasingconstΓuctingqualityandsafety.Atthesametime,constΓuctionsitesaΓeoneofthemostchallengingenviΓonmentsfoΓautomationdueto

constantlychangingteΓΓain,layouts,andhumanactivity.

AustralianroboticscompanyFBRsHadrianXaddΓessestheseconstΓaintsbyautomatingoneofthemostlaboΓ-intensivetasksinconstΓuction:stΓuctuΓalwallbuilding.HadΓianXisanautonomous,mobileconstΓuctionΓobotthatusesaΓoboticaΓmmountedonavehicleplatfoΓmtoplaceconcΓeteblocks.TheΓobothasbeenpilotedonanactiveconstΓuctionsiteintheUS,andhasdemonstΓatedtheabilitytoconstΓuctstΓuctuΓal,load-beaΓingwallswithinaday.5

BostonDynamicsandFieldAIaΓetacklingadiffeΓentbottleneck:sitemonitoΓingandinspectioninconstΓuctionenviΓonments.ConstΓuctionsitesaΓedifficulttomonitoΓconsistentlyduetochangingconditionsandsafetyΓisks,makingdatacollectionlaboΓ-intensiveandeΓΓoΓ-pΓone.ThepaΓtneΓshipcombinesBostonDynamics’SpotquadΓupedΓobotwithFieldAl’sFieldFoundationModelstoenableautonomousinspection,mapping,andmonitoΓing.AlΓeadydeployed

acΓossmultiplelocations,thesolutionsuppoΓtsfleet-levelautonomyandcooΓdinatedopeΓation,andhasdeliveΓedoveΓ90%Γeductionsininspectionanddocumentationtime,eaΓlieΓissuedetectionthatΓeducesΓewoΓkcosts,andimpΓovedwoΓkeΓsafety.6

Automationisbecomingincreasinglycriticalinagricultureaslaborshortagesintensifyinmanyregions.7HoweveΓ,scalingautomationinagΓicultuΓe

ΓemainschallengingduetothehighlyvaΓiablenatuΓeoffaΓmingenviΓonments一wheΓelighting,teΓΓain,andcΓopvaΓietiesdiffeΓwidelyacΓossfields一

andtheΓelianceonheteΓogeneousfleetsofmachines,includingtΓactoΓs,haΓvesteΓs,andspΓayeΓs.TorqueAGI,aUS-basedstaΓtupbuildingfoundationmodelsfoΓΓoboticautonomy,addΓessestheseconstΓaintswithphysics-infoΓmedAlfoundationmodelsthatcanhandledensefoliage,iΓΓegulaΓplant

geometΓy,andmultimodalpeΓception,whileopeΓatingacΓossdiffeΓentmachines.ToΓqueAGliscollaboΓatingwithJohnDeeretoadvanceAlfoundationmodelsforthenextgenerationofintelligentagriculturalrobots.8

Illustrativecases

Continuedonnextpage

CapgeminiResearchInstitute2026

17

PhysicalAI:Takinghuman-robotcollaborationtothenextlevel

Industries

Illustrativecases

Healthcare/eldercare

Wandercraft,aFranceandUS-basedroboticscompany,isdevelopingAIpoweredmedicalexoskeletonsthatenablepeoplewithspinalcordinjuries,

stroke,andotherseveremobilityimpairmentstostandandwalk.ItslatestdevicethePersonalExoskeletoniscurrentlyinclinicaltrialsandisdesignedforeverydayindoorandoutdooruse.ThedeviceusesAIforbalanceandmovement,adaptingcontinuouslyinrealtimetosupportstablewalkingacrossvariedsurfacessuchasconcrete,carpet,andtile.9

ElliQ,anAI-poweredcompanionrobotforolderadultsdevelopedbyIntuitionRobotics,isbeingintroducedtoJapanthroughapartnershipwith

JapanesetradingcompanyKanematsuCorp.ThecollaborationtargetsJapansrapidlyagingpopulationandtheresultingshortageofcaregiversandnursinghomestaff.ElliQproactivelysupportsolderadultswitheverydayneeds,includinghealthmanagement,preventivecare,communication,

monitoring,andsocialandcognitiveactivities.10

Energy

AI-enabledrobotsfromUS-basedLuminousRoboticswereusedtohelpinstallnearly500,000solarpanelsatENGIEs250MWsolarfarminVictoria,Australia.LuminoussLUMIrobotsautonomouslyliftandplacepanelsontomountingstructuresusingAI-drivenpick-and-placesystems,while

humancrewscompletefinalfastening.Thisreducesheavymanuallabor,improvessafety,andincreasesefficiency.Therobotsdemonstratedahighdegreeofflexibility,operatingeffectivelyacrossarangeofweatherconditions.Morebroadly,automatingsolarconstructionisexpectedtolowercostsandspeedupconstruction,enablinglargerscalesolardevelopments,whilereducingtheneedforhumanlaborinremoteandinhospitable

outdoorenvironments.11

Sources:Informationcompiledfrompubliclyavailablesecondarysources.

AprofessorofroboticsataUK-baseduniversitysays:"Traditionalrobotsareoptimizedtoexecutepredefinedmotions,withlimitedunderstandingofintentorrealworld

impact.PhysicalAIfundamentallychangesthisbyenablingrobotstoperceivetheirsurroundingsandreasonaboutcontext.Indoingso,itopensupproblemdomainsthathaveresistedautomationfordecades–preciselybecausetheyrequireunderstanding,notjustexecution.”

CapgeminiResearchInstitute2026

18

PhysicalAI:Takinghuman-robotcollaborationtothenextlevel

Evolutionofroboticintelligence

Theevolutionofintelligenceinrobotics(ataglance)

Late2010s–early2020s

Domain-specificLearning-enabled

intelligentrobotsrobotics

Early–mid2000s

1990s

2010s

Pre-programmedautonomy

Today

1960s–1980s

Sensor-aware

robots

PhysicalAIera

Hard-codedautomation

Robotsbegintolearnfromexperienceandsimulation

Robotsasadaptive,autonomoussystems

Robotsexecutefixed,

pre-programmedmotions,

butwithnoawarenessof

theirenvironment

Robotsgainlimitedawareness

oftheirsurroundings,but

decision-makingisstill

rule-based

Robotsoptimizedfornarrow

tasks(enabledbyadvancesin

AIandperception)

Robotsoperateindependently

incontrolledsettings(enabled

bysensingcombinedwith

onboardsoftware)

•ASIMO(2000):Honda’shumanoid

ΓobotdemonstΓatedwalking,balance,andbasicinteΓaction

•EaΓlyseΓviceΓobotsinΓeseaΓchanddefense

•Kivarobots(laterAmazonRobotics):AutonomouswaΓehouselogistics

•CollaboΓativeΓobots(cobots)onfactoΓyflooΓs

•SuΓgicalandinspectionΓobotsinhealthcaΓeandinfΓastΓuctuΓe

•Vision-guidedindustΓialΓobots

•FoΓce-contΓolledaΓmsfoΓpΓecisionassembly

•EaΓlymobileΓobotsinwaΓehousesandlabs

•DeepleaΓningfoΓpeΓceptionandmanipulation

•Cloud-connectedΓobotfleets

•Sim-to-ΓealtΓainingpipelines

•Roboticfoundationmodelsenabling

geneΓalizationacΓosstasksandtheabilitytoopeΓateinunstΓuctuΓedenviΓonments

•Multi-ΓobotandAlagentoΓchestΓation

•RapidpΓogΓessinhumanoidandgeneΓal-puΓposeΓobots

Keymilestones

•Unimate(1961):ThefiΓstindustΓialΓobot,deployedonGeneΓalMotoΓs’assemblyline

•EaΓlyindustΓialaΓmsinautomotivemanufactuΓing

•PLC-dΓivenautomationandfixedpΓoductioncells

CapgeminiResearchInstitute2026

19

PhysicalAI:Takinghuman-robotcollaborationtothenextlevel

Dr.MiladMalekzadeh,Co-FoundeΓandVicePΓesidentAlatNeuΓaRobotics,aGeΓmany-basedΓoboticsfiΓm,says:

“PhysicalAIwillbeusedacrossvirtuallyeverytypeofcontext–fromindustrialsettingstomedical,home,andserviceenvironments.Progressisbeingdrivennotonlybyadvancesinindividualmodels,butincreasinglyby

platformapproachesthatmakeiteasiertocombine,reuse,anddeployintelligenceacrossdifferentrobotsandusecases.”

20

PhysicalAI:Takinghuman-robotcollaborationtothenextlevel

Adiscussionwith

DeepuTalla,

VPandGM–Robotics&EdgeAI,NVIDIA

WhatΓecentadvancesaΓedΓivingtheinflectionpointinphysicalAI?

Inthelast12to24months,twotechnologieshavereachedameaningfullevelofmaturity,bringingusintowhatI

wouldcallagoldenageforphysicalAIandrobotics.First,

foundationmodels–nowextendingbeyondlanguageto

visionandaction–makeitpossibletomovebeyondbrittle,task‑specificsystemstowardmoregeneral‑purposerobotbrains.Second,advancesinsimulation.Thegapbetween

simulationandtherealworldhasnarrowedformanyuse

cases,allowingrobotstobetrainedandtestedextensivelyinsimulation.Thisisordersofmagnitudefaster,safer,andcheaperthanreal-worldtesting,compressingdevelopmentcyclesthatoncetookyearsintohoursorevenminutes.

WhydoesphysicalAIΓepΓesentsuchasignificanteconomicoppoΓtunity?

Themajorityofglobaleconomicactivityistiedtophysicalindustries.

温馨提示

  • 1. 本站所有资源如无特殊说明,都需要本地电脑安装OFFICE2007和PDF阅读器。图纸软件为CAD,CAXA,PROE,UG,SolidWorks等.压缩文件请下载最新的WinRAR软件解压。
  • 2. 本站的文档不包含任何第三方提供的附件图纸等,如果需要附件,请联系上传者。文件的所有权益归上传用户所有。
  • 3. 本站RAR压缩包中若带图纸,网页内容里面会有图纸预览,若没有图纸预览就没有图纸。
  • 4. 未经权益所有人同意不得将文件中的内容挪作商业或盈利用途。
  • 5. 人人文库网仅提供信息存储空间,仅对用户上传内容的表现方式做保护处理,对用户上传分享的文档内容本身不做任何修改或编辑,并不能对任何下载内容负责。
  • 6. 下载文件中如有侵权或不适当内容,请与我们联系,我们立即纠正。
  • 7. 本站不保证下载资源的准确性、安全性和完整性, 同时也不承担用户因使用这些下载资源对自己和他人造成任何形式的伤害或损失。

最新文档

评论

0/150

提交评论