AOCS/GNC Autonomy and FDIR Airbus challenges and way forward
Challenges in Achieving Long-term Autonomy
Transcript of Challenges in Achieving Long-term Autonomy
Challenges in Achieving
Long-term Autonomy Hai Lin, Discover Lab
Electrical Engineering Department University of Notre Dame
E-mail: [email protected]
q Thekeydifferenceisthatautonomoussystemsneedtoworkin“Open”realworldthatishighlyuncertainanddynamic–needperception,reasoning,andadaptation.
q Long-termmeansthattheoperationtimeismuchlongerthanthevalidityofaprioriassumptions/information
What we mean by long-term autonomy?
Complexity
Uncertainty
Smallsearchingspace→Largesearchingspace
ScalabilitySinglesystem→Multiplecomponents→Distributeddesign
Static/knownenvironment
SimpledynamicsSimpletask
High-dimensionaldynamicsComplexmissions
UncertainDynamicenvironment
• InDiscoverLab(DistributedCooperativeSystemsResearchLab)atNotreDame,weareusingmulti-robotsystemsandhuman-machinecollaborationasworkingexamplestostudythedesignprinciplesforengineeredcomplexsystems.
• Weaskhowtodesignintelligentphysicalsystemswithprovablycorrectperformanceeveninuncertain,dynamicenvironments.
DISCOVERLab at Notre Dame
Combine top-down and bottom-up design
Local motion control ler s design
...
Global mission
DecompositionTop-dow n Mission plan N
Local motion planning
Mission plan 1
Bottom-up
Motion plan NMotion plan 1...
Motion control ler 1
Motion control ler N...
Not feasible
...
Global mission
DecompositionMission plan NMission plan 1
Synergy level...
Not feasible
...Pr imitives
Synergy level
Pr imitives
Environment
Top-dow n
Bottom-up
No
No
...
KMI
DecompositionKNMIK1MI
Assume-guarantee Compositional Ver i f ication
YesCounterexamples?
Modify Assumptions?
Success
Yes
Top-down Task Decomposition
Dai,J.,Benini,A.,Lin,H.,Antsaklis,P.J.,Rutherford,M.J.,&Valavanis,K.P.(2017).Learning-basedformalsynthesisofcooperativemulti-agentsystems.arXivpreprintarXiv:1705.10427.
• Giventeammissionasformalspecifications
• Needtodecomposetheglobalmissionintoindividualrobottasks.
• Automaticreconfigureifthesituationchanges
Integrated Task and Motion planning
daSilva,R.R.,&Lin,H.(2018).ScalableIntegratedTaskandMotionPlanningfromSignalTemporalLogicSpecifications.arXivpreprintarXiv:1803.11247.
Human-machine collaboration Distinctfromthemajorityofexistingwork,whichfocusedonhuman-machineinterface,orhuman-awaremotionplanning,weareinterestedinq Missionplanning:Human-machinecollaborationtoachieve
high-levelmissionsinaprovablycorrectmannerq Integratedtaskandmotionplanning:jointmanipulation
§ Wu,B.,Hu,B.,&Lin,H.(2017).ALearningBasedOptimalHumanRobotCollaborationwithLinearTemporalLogicConstraints.arXivpreprintarXiv:1706.00007.
§ Zheng,W.,Wu,B.,&Lin,H.(2018).POMDPModelLearningforHumanRobotCollaboration.arXivpreprintarXiv:1803.11300.
Post-modern control era
MachineLearning
ComputerSciences
ControlTheory
Formal methods Intelligent Control
Artificial Intelligence
Uncertainty:Notjustdisturbancesornoises,Uncertaintiesareeverywhereduetounanticipatedanddynamicallychangingreal-worldsituationsComplexity:Complexmissionsthatmaynotbecapturedbyoptimization,stabilityorregulationproblemsScalability:Beyondsingle-loopfeedbackcontrol