Lectures TDDD10AIProgramming PuttingItAllTogether CyrilleBerger 1AIProgramming:Introduction 2Introductionto 3AgentsandAgents 4Multi-Agentand 5Multi-AgentDecision 6CooperationAndCoordination 7CooperationAndCoordination 8MachineLearning 9AutomatedPlanning 10PuttingItAllTogether 2/65 Lecturegoals LecturesSummary Learnhowallthealgorithmsand conceptsfittogether 3/65 4/65 Lecturecontent AgentArchitecture CommunicationRelay Trafficsurveillance Sensing AgentArchitecture Stream-BasedReasoning DetectingandTrackingcars Multi-agentsensing Rescueoperation HumanDetectionand Planning Delegation BuildingScanning Masterthesis 5/65 ArchitectureTheory UAVArchitecture 7 8 CommunicationRelay Communicationisakeycomponent ofmulti-agentsystems Communicationcanbe CommunicationRelay Communicationnetworkscanbe disabledordestroyed Forinstance,during9/11,cellularphoneswerenotworking 10 CommunicationRelay:visibility CommunicationRelay:multiplepath 11 12 Networkbuildingwithdelegation CommunicationRelay ThegoalistofindachainofUAVthatminimizethenumberof Thisisacomplicatedoptimization Getworseifyouhavemultiple UseVoronoigraphandTreebased Network buildingmission Sequence Deploy basestation Sequence Flyto Thisworkin2D,butwhatabout Grab Deploy relaystations Concurrent Flyto Grab Deploy relaystation1 Sequence ... 13 14 Howtogetfullcoverage? Findinganoptimalcoveragein3D isunreasonable Instead,whenanagentisinan areawithlowcommunication,it shouldsendarequestforgetting arelaystationinstalled Trafficsurveillance 15 TrafficsurveillanceScenario(1/2) TrafficsurveillanceScenario(2/2) Continuouslygatherinformationfrom manydifferentsources. Selecttherelevantinformationforthe currenttask. Derivehigher-levelknowledgeaboutthe environmentandtheUAV. Suchasdetectingmisbehavingdrivers Correctlyinterpretwhatisgoingon. CoordinationofUAVswithother CoordinationofUAVswithother humanpoliceforces(calledmix initiative) 17 Sensing 18 Stream-BasedReasoning TheSense-ReasoningGap Stream-BasedReasoning Autonomoussystemsproduceand processsequencesofvalues incrementallycreatedatrun-time. Thesesequencesarenaturalto modelasstreams. Stream-basedreasoningis incrementalreasoningover streams. Stream-basedreasoningcaptures thecontinuousreasoningwith minimallatencynecessaryto reacttorapidchangesinthe environment. 21 IncrementalEvaluationofTemporalLogicalFormulas 22 StateStreams Thesemanticsoftheseformulasisdefined overinfinitestatesequences.Progression isonetechniquetocheckwhetherthe currentprefixissufficienttodeterminethe truthvalueofaformula. 23 24 TrafficMonitoring DetectingandTrackingcars 26 Anchoring(1/3) Anchoring(2/3) Theobjectiveoftheanchoringprocessisto connectsymbolstosensordataoriginatingin thephysicalworldsothatthesymbolsrepresent theobjectsintheworld 27 28 Anchoring(3/3) ChronicleRecognition 29 30 MultiUAVTrafficMonitoring Multi-agentsensing 32 DyKnowFederation Cooperation whathappenwhenanhelicopterneedto leaveitspatrolareatogoinpursuitof anoffender? Either,needtofindareplacementforpatrollinghisarea Oranotherhelicoptertopursuetheoffender Surveillance Sequence Surveillance Sequence Pursuit Sequence 33 34 RescueoperationScenario(1/2) Rescueoperation 36 RescueoperationScenario(2/2) Explorationtofindallvictims Divisionofareaintoscanningarea Taskallocation Detectionofvictims HumanDetectionandApplication Rescueof Taskallocation 37 VictimsDetection Leashing 39 40 HowtoconstructaTask-SpecificationTree? Planning 42 Naturaldisasterexample PlantoTask-SpecificationTree Anautomatedplannertakesa problemdescription Example:Supposetherehasbeen anaturaldisaster Objectives Availableresources,actions,… Objective:100peopleshouldhavefood, medicineandwater WehaveasmallfleetofUAVsavailable Howtodescribethistoaplanner? Howtogeneratea …andgeneratesaplanthatachieves theobjectives 43 44 Naturaldisasterexample:GeneralWorldDescription High-leveldescriptionoftheworld weoperatein Naturaldisasterexample:ProblemDescription Thecurrentstateoftheworld Locationsofinjuredpeople, availabilityofcrates,… Entities:UAVs,crates,people,… Properties:Entitieshavelocations,UAVshave capabilities,… Agoaltobeachieved (and(forall?person(has-crate?personfood)) (has-crateperson2medicine)(has-crate person7medicine)…) Actionsthatcanbeperformed (:operator(deliver-crate?uav?crate?from?to) (:precond(and(has-capability?uavcarry-crates) (at?uav?from)(at?crate?from)…) (:phase:duration(flight-time?from?to) :effects(:assign(location?crate)?to) (:assign(location?uav)?to)…)…) 45 46 PlanningtoTask-SpecificationTree Resultingplan Theresultingplansshouldbeabletoexpress: Concurrency:sequentialplannersarenotapplicable Precedence:uav7picksupcarrier2afteruav2loadscrates Lackofprecedence:Onlywaitforotheragentswhenyouhave Approximatetiming:(exactdurationsareunknown) to! 47 48 Delegation(1/2) Missionconsistingofaflight action+agoaltoachieve Delegation 50 Delegation(2/2) DistributedPlanning Verifyexecutabilitythroughon-boardfunctionalitiesduring planning Motionplanner Scheduling,resourcereasoning,constraintreasoning Infeasibleactionimmediatebacktracking! Useofaconstraintsolver 51 52 DelegationExample BuildingScanning 53 BuildingScanningpart1 BuildingScanningScenario 55 56 BuildingScanningpart2 LecturesSummary 57 58/65 Masterthesis InvariousdomainofArtificial Intelligence Masterthesis Planning KnowledgeRepresentation MachineLearning Robotic Sensing 60 Objectdetectionandrecognition GuidedExploration Lookingforvictims,usingpriorknowledgeaswellasnew observation Objectrecongitionusing deeplearning Prioritizedscanarea Implementationwitha groundrobot Evaluationonaerial 61 62 HumanRobotInterractions Voicecommands Poserecognition Andmanymore... 63 65/65