b'SUPERCOMPUTING FOR BETTER COMMUTINGIN PURSUIT OF FUEL ECONOMY, MOBILITY Tom Karnowski, shown here with the Summit supercomputer, is leading a team of scientists using high performance computing to improve mobility and increase fuel efficiency in an By Stephanie Seay HPC4Mobility project at ORNL. Photo credit: Carlos Jones, Oak Ridge National LaboratoryInaprojectleveragingcomputervision,The fuel savings potential is substantial. TheGRIDSMARTisexcitedtobeworkinglearningonORNLssupercomputers.resourcesmadeavailablethrough machine learning, and sensors, Oak RidgeUSDepartmentofEnergyestimatesthatwith ORNL on this project, said Jeff Price,Reinforcement learning basically teaches aHPC4MobilitycanenableU.S.industry National Laboratory scientists are workingidling from heavy- and light-duty vehiclesGRIDSMARTchieftechnologyofficer.computer how to play a game without beingtooptimizeenergyefficiencyandreduce withprivatecompanyGRIDSMARTcombined wastes about 6 billion gallons ofMultimodalurbanmobilityposesveryexplicitly programmed to do so. In this case,emissions, said Claus Daniel, Sustainable Technologies,Inc.todemonstratehowfuel per year. As vehicles idle at intersectionscomplicatedchallenges.Bringingtobearthe game is saving fuel while not sacrificingTransportationprogramdirectorand stop lights can be programmed to improveand other locations, fuel is wasted. ORNLcapabilitiesinhigh-performancethroughput, Karnowski said. HPC4Mobility program lead at ORNL. Were fuel economy and reduce emissions whileTheORNL/GRIDSMARTprojectwasonecomputingandmachinelearningonTheprojectleverageshighperformanceworking hand-in-hand with a private partner facilitating the smooth flow of traffic.of the first awarded funding under the DOEGRIDSMARTs unique data and large installcomputingsystemsattheOakRidgeto leverage DOEs computing resources and GRIDSMARTtrafficcamerasarealreadyVehicleTechnologiesOfficesnewHighbase will provide some fascinating insights. Leadership Computing Facility, a DOE Officedeep learning expertise to solve a real-world beingusedbycommunitiesaroundthePerformance Computing (HPC) for MobilityThe first phase for ORNL researchers hasof Science user facility at ORNL, such asmobility challengeone that will save energy world to replace in-road sensors traditionally(HPC4Mobility) program. As part of the HPCbeen to take images from GRIDSMARTstheSummitsupercomputertheworldsand improve traffic flow.used to detect vehicles and inform trafficfor Energy Innovation Initiative, the programoverheadtrafficcamerasandcomparemost powerful openly accessible computer.TheDOEVehicleTechnologiesOffices signal timing. These smart cameras providebringstogetherthesupercomputingthemtoground-levelphotostocreateaThesystemswillbeusedtoperformEnergyEfficientMobilitySystems(EEMS) a real-time, birds eyeresourcesanddatabase.Theresearcherswanttotrainsimulationsofintersectionsandcomeupprogram,apartoftheagencysOffice view of intersections,scientificexpertisethe cameras to estimate the fuel economywith mathematical strategies to guide trafficofEnergyEfficiencyandRenewable gatheringdatathatIDLING VEHICLESofDOEsnationalof various vehicles present at intersectionslight timing.Energy,providedfundsforthisproject. can guide timing andlaboratoriesinby categorizing their size and vehicle class.Computational resources were provided by traffic flow strategies.WASTE ABOUT 6partnershipwithThewholeideaistoteachcamerastotheDOEAdvancedScientificComputing BILLION GALLONSindustrytofindORNLresearchersdevelopedasystemestimate fuel consumption and then teachResearchprogramthroughtheASCRStephanieSeayisasciencewriter Thegoalsofthesolutionstoreal- to combine the resulting photos with fuelan entire grid of those cameras to manageLeadership Computing Challenge.atOakRidgeNationalLaboratory, programaretoOF FUEL PER YEAR.worldtransportationconsumptiondatafordifferentvehicletraffic lights to make the system more fuelresearchingandwritingcontentto teachGRIDSMARTenergy challenges.classes, which resulted in a rich dataset ofefficient, Karnowski said.ORNLismanagedbyUT-Battellefortheelevatethegroundbreakingworkof cameras to estimatelabeled images. Department of Energys Office of Science,thelabsscientistsandengineers. the fuel efficiency of vehicles at intersectionsCreatinganintelligenttransportationORNL collaborators on the project includethe single largest supporter of basic researchPreviously,shewassenioreditorat AnymachinelearningprojectwillonlyTravisJohnston,ThomasNaughton,WaelmarketintelligencefirmS&PGlobal, and then to control traffic signal timing insystemintraffic-denseurbanareasinthephysicalsciencesintheUnitedwhere she covered the energy sector as order to save energy while optimizing trafficrequiresobservationbeyondhumanbe as good as the data youre putting in,Elwasif, Jonathan Sewell, Russ Henderson,States. DOEs Office of Science is workinga business journalist.throughput,explainedprojectleadTomcapacity, and the amount of data generatedKarnowski said.and Husain Aziz. toaddresssomeofthemostpressing Karnowski of ORNLs Imaging, Signals, andby the GRIDSMART cameras makes it anThe second phase of the project is to create aThis project is exemplar for how nationalchallenges of our time. For more information, Machine Learning Group.excellent fit for the HPC4Energy program,software application by using reinforcementlaboratoryhighperformancecomputingplease visit science.energy.gov. nKarnowski noted. 31 32'