1、使用BAETYL构建边云融合的智能应用李乐丁 百度智能云物联网主任架构师|01关于边缘计算关于边缘计算需求理念方法02技术和创新技术和创新云原生人工智能03合作与应用合作与应用社区商业04未来发展未来发展规划参与目录目录CONTENT|关于边缘计算你好01|计算无处不在从中心到边缘|Cloud IDCCDNCDN5G/MEC5G/MEC5G/MEC5G/MEC5G/MEC110ms10100ms100msThing EdgeLAN EdgeWAN EdgeIoTIoTIoTIoTIoTIoTAPPAPPAPPAPPCDN边云融合云上开发，边缘处理，数据上行，知识下行|Edge SideClou

2、d SideEdge FrameworkEdge node InfraSynchronzationData ProcessingConn&StorageMgmt.ConsoleCloud InfraAI Models&AlgoApplications&ConfDevices&HDWData UploadingKnowledge DownloadingoApplicationoML ModeloDevice ControloetcVideoCamera&ScreenAudioMicphone&SpeakerDataSensor&MachineoLog&DiagnosticoTraining Sa

3、mpleoRuntime Statusoetc百度智能边缘体系边云融合解决方案|Baidu IntelliEdgeAPPData ingressRegion EdgeDedicate ServersX86,ClusterLocal EdgeIndustry GatwaysARM,X86Device EdgeDrone,Robots,etc.ARM,RISC-VBaidu Cloud-Edge Fusion ConsoleRemote ManagementLocal agentData managementReg&ActivationLocal repository(IoT,image,file

4、s)Local sysSched.HardwareAbstractLifecycleAPIOPC-UAModbusServerless compMQTTRTSPMQFaaSML InferVideoUser appsData analysisDevice mgmtData ExchangeLaunchCLIOSStorageCloud ClusterTerminal Devices,Gateway,TTU,Meters,Camera,etc.Local conn,Local proto3rdpartyservicesUnifiedVarious OSVarious chipsetsVariou

5、s NetworkFusionCloud mgmt.,Edge deployCloud develop,Edge execData upload,Know downOpenCustomized applicationsOpen sourceCloud NativePracticalFull stack data analysisFull stack edge CV solutionBaetyl，通用边缘计算框架LFEdge项目|qLaunched in 2017 as Baidu IntelliEdge,open source in 2018qJoin LFEdge umbrella in 2

6、019qProvide Cloud Native environment for all applications and AI servicesqUnified management for both cloud and edgeqUnified security model and data privacy protectionqVendor neutral,Apache Licensehttps:/baetyl.io/技术和创新你好02|2019，Baetyl 1.0架构以容器为中心|o Leverage Dockero Edge node,single nodeo High compa

7、tibilityo x86,x86_64,arm32,aarch64o Linux,Windows,macOSo Integrated with serviceso Local MQTT brokero Local Function serviceo Remote msg synchronizero Proprietary Managemento with Open Protocol2020，Baetyl 2.0架构释放Kubernetes的力量|o Leverage Kuberneteso Edge cluster,multiple nodeo Open-source managemento

8、 baetyl-cloudo with OpenAPIo Self upgradableo Baetyl is running as a Podo More integrated serviceso Local KV storageo Logging&Tracingo Remote metrics report开放生态，厂商中立最大化兼容性|qA Cluster-to-Cluster designinspired by K8S Federationself-schedulinglocal load balance&failoverqNo*modification*to kubeletno sp

9、ecified K8S vendor requirementno specified K8S version requirementqFull K8S semantics in edge instancePod,Deployment&ServiceDaemonSetConfigMap&SecretSingle NodeBaetylRancher K3SApplicationCloud SideAWS EKSBaidu CCEBare-metal K8SMultiple NodesBaetylOrig.K8SApplicationNodes on IaaSBaetylIaaS K8SApplic

10、ationApp.AApp.BApp.A1 replcaApp.B1 replcaApp.B3 replicaApp.A3 replicaBaetyl与百度AI的集成自动模型转换|SourceDeviceTargetAnyIntel MovidiusOpenVINOPaddlePaddleCUDAPaddleLiteTensorFlowEdgeTPUTF-LiteTrain AI modelon cloudInferenceon edgeMatchwith hardwareConvertReport hardwarecapbilitiesAccelerationBIE视觉解决方案一站式升级，从

11、普通摄像头到智能摄像头|Steps:1.Power on2.Connect camera to POE line3.Connect AI chipset to USB portWorkflow:1.Load AI model2.Sampling video as images3.Do inference,accelerated by AI chipset4.Save target image5.Send result to next process node6.Push target image into cloud合作与应用你好03|Baetyl与EdgeX Foundry的合作LFEdge

12、社区互助|Shanghai DivisionBeijing DivisionShenzhen Divisionq EdgeX Challenge China 2021q An international hackathon event on EdgeX,hosted by LF Edgeq Consists of tracks of commerce&health,industrial&logisticsq With Special Baetyl AwardBIE智能电网解决方案电力能源的智能调配|BaetylApplicationsSyncAI platformBaetyl CloudApp

13、licationsDevicesModelsDeployScheduleCloudEdgeDeviceMgmt.OS&K8SDroneRobotsMetersMoreIoT SDKModbus/OPC/RTSP etcThe AI platform includes:Cloud management suite,responsible for batch distribution and configuration update of AI applicationsLocal client,supports deployment on the edge and end,and is respo

14、nsible for AI application start-stop and state management at the edge layer and end-sideEdge service includes:When the local client is deployed on the edge node,the cloud delivers the application to the edge layer,the edge layer collects the terminal data of the device layer,implements prediction an

15、d inference on the edge node.When the local client is deployed on the terminal device,it is applied to the terminal under the cloud,and prediction and inference are realized on the terminal.BIE智能质检方案迈向工业4.0|Model deployment time is longTraining data collection is complexSpare parts quality inspectio

16、n cloud inference delay affects quality inspection speedModel deployment went from 1 day to 1 minuteAutomatic collection of training data and automatic upload to the cloudEdge AI inference realizes edge quality inspection and improves quality inspection efficiencyIssuesEffectsManufactoryCamerasCPU&G

17、PUBaetylVideoSamplingAIModelMarks&SaveInspection Machinelightning WireCellphoneBaidu CloudPaddlePaddle工工业业质质检检应应用用平平台台False resultRaw imagesAI resultDataNew modelsNew configurationsV2X自动驾驶5G和智能道路升级|5G Base StationMEC NodeMEC Node5G Base StationAutopilot carRoadside PerceptionMEC ClusterBaetyl with A

18、IV2X AlgorithmMEC NodeV2X Remote ConsoleKey data uplinkVideo files uplinkModel upgradeConfiguration upgradeCross-region V2XBaidu IoT PlatformOver 60%reduction in automatic driving costs to help solve cross-regional road synergy challengesGlobal optimization of a wide range of multi-base station node

19、s未来发展你好04|Baetyl 3.0云与边缘的彻底融合|qDeep integration with K8SqAll resource into edge,even Namespace!qCRD from cloud to edgeqUniversal managementqBaetyl-cloud as a CRDqJust kubectl all workloads for cloud&edgeqComing soon!qpreview Baetyl 2.6 in June.2022加入我们用科技让复杂的世界更简单|https:/baetyl.io/https:/ Cloud IoThttps:/