S&AS:INT:AutonomousMulti-RobotVisualMonitoringforUrban,Agricultural,andNaturalResourceManagement
PI:A.Roy-ChowdhuryCo-PIs:K.Karydis,Q.Zhu,A.Mourikis,G.Jenerette
SeniorPersonnel:N.Abu-GhazalehUniversityofCalifornia,Riverside
Research Objective
Framework for multi-agent aerial imaging systems that can adapt autonomously toenvironmental conditions, technical constraints, and ethical considerations for dataanalysis, trajectory planning, and computation/communication resource management,with applications in monitoring vegetation health.
Overview:Thisworksynthesizesideasfromvisualsensing,autonomousrobotnavigation,decisionmaking andsystemoptimization todevelopageneralframeworkforadaptivemulti-agentvisualmonitoringappliedtonaturalresourcemanagement.
A. AdaptiveVisualAnalysis1. Adaptivealgorithmdesignandselection2. Resource-constrainedoptimaldataprocessing
B. AutonomousAerialRobotNavigation1. ReliableGPS-deniedposeestimationwithmulti-modalvisualinput
a. Useofmulti-modalvisualinputb. Useofsatellitemapsforbounded-uncertaintyvisuallocalization
2. Trajectoryplanningandcontrolunderconflictingconstraintsa. Aerodynamicsandenergeticsofroboticflightinrealisticoperationalconditionsb. Sensory-basedadaptiveplanningunderconflictingconstraints
C. System-levelDecisionMakingandAdaptation1. Sensingandnavigationco-adaptation2. Computationandcommunicationconstraintsanalysisandmitigation
Major Research Tasks
A. Online Video Summarization: Video Fast-Forwarding via Reinforcement Learning (CVPR 18)
Goal: Develop a method for fast-forwarding through a video that is computationallyefficient, causal, online and results in informative video segments
Approach:
• Formulate fast-forwarding as Markov Decision Process (MDP)- Solve using Q-learning• Online framework to deal with incremental observations without requiring to store and
process the entire video• Experiments on 2 standard summarization datasets – about 6%-20% improvement in
coverage with 80% reduction in number of frames processed
ReinforcementLearningFrameworkforFast-ForwardingVideos
Approach:
• Introducefaithfulmodelstocaptureandmitigatetheimpactofvariousconstraints,e.g.,energyconsumptionatvariousspeedsorincloseproximitytootherrobots.
• Introduceaccurateandreliablevisual-inertialodometrymethodstoclosefastaction-perceptionloopsforautonomousnavigation.
• Developnewmulti-robotdeploymentalgorithmsthatareawareofenergyconsumptiontooptimizerobotformationsandtaskallocation.
• Developresourcemanagementpoliciestotradeoffaccuracyandperformancetomeetcomputationalconstraintsandoperateundervariableconnectivity.
B/C. Autonomous Navigation under Energy, Computation and Communication Constraints
Goal: Develop and evaluate multi-robot motion planning and navigationalgorithms under energy, computation and communication constraints
Quadrotorflightisfoundmoreenergyefficientwhenflyingforwardat5-8m/s.
Evaluation: Imaging and Modeling Vegetation
Goal:Identifyandcharacterizevegetationcanopyforrapidevaluationofhealthandproductivity.
Approach:
• CombinefieldsurveyswithUAV-basedimageryinvisibleandthermalwavelengths• Rapidlyidentifyspeciesofdominantplants,andtheirhealth• Usestructure-through-motionapproachtoreconstruct3-dimensionalgeometryofvegetation
Initialcollectionofhighresolutionimageryfromacommunitywithtwodominantplantspecies.
Exampleof3-Dreconstructionoffieldsiteusingthestructure-throughmotionapproach.