Outline Technical Goal • Technical Goal & Motivation Demonstrate autonomous control of an interferometer Autonomous Sequencing and • Basics of Optical Interferometry instrument by implementing a model-based fault protection Model-based Fault Protection for system (Livingstone) and autonomous sequencing (Exec) on • Background on Remote Agent Space Interferometry a ground-based interferometer testbed, in the context of a • System Architecture representative observation scenario. Michel Ingham, Brian Williams Thomas Lockhart, Amalaye Oyake, • Livingstone: Model-based MIR MIT Space Systems Lab Micah Clark, Abdullah Aljabri • Executive: Autonomous Sequencing MIT Artificial Intelligence Lab Caltech Jet Propulsion Lab Cambridge, MA Pasadena, CA • Lessons Learned • Future Work i-SAIRAS June 21, 2001 Chart: 1 Chart: 2 Chart: 3 Motivation Optical Interferometry Remote Agent Acquire starlight (L & R) • Model-based autonomy has • sophisticated monitoring and control s/w • slew Siderostat to estimated significant potential for NASA target angle • AI technology used to encode operational rules and system constraints missions • with Siderostat tracking, within flight s/w perform Star Tracker search • Learn to represent system • Star Tracker locks onto target • ground operators rely on RA to monitor s/c and achieve mission goals star engineering knowledge of a Stellar Wavefront Acquire fringe External • flight validated on DS-1 in May ‘99 complex instrument: SIM Delay • calibrate PZT Dither α – numerous components • homeset Delay Line, zero Mission • 3 primary modules: Scripted Baseline Fiducial Laser Counter, lock Internal Manager Executive Fiducial – significant component interaction Metrology – Planner/Scheduler Beam – multiple failure modes Combiner Siderostat • slew Delay Line to estimated – Smart Executive Siderostat Delay delay position Internal – multi-step recoveries Model- – Livingstone model-based MIR Metrology Line Planner/ • with Delay Line tracking, based Laser Source Scheduler perform fringe search MIR • Successful ground test-bed • Fringe Tracker locks onto demonstration: 1 st step toward Fringe fringe Detector broader acceptance Perform science integrated with interferometry RTC s/w TPF measurement Chart: 4 Chart: 5 Chart: 6 Page 1
Livingstone: Livingstone: System Architecture Model-based MIR Model-based MIR • Livingstone provides the following capabilities: • Use common-sense behavioral models of spacecraft components & CORBA MESSAGES CLASH MESSAGES REMOTE AGENT � Monitoring of predicted vs. observed state (consistency checking) subsystems � Mode identification in case of discrepancy • Transition system formulation, based on qualitative representations � Reconfiguration suggestions over finite domains: Commands counter-delta = { hold, track, slew, out } RTC EXEC • Performs significant deduction in the sense/response loop • Represent dynamics with probabilistic automata ACE-TAO ORB Controller Model Goals • Models compile to propositional logic: Telemetry Data State Recovery Commands Internal Metrology Internal Metrology Recoveries updates Requests Component Model Component Model mode = locked ⇒ (not (counter-delta = out)) s’(t) mode mode Mode constraint: Telemetry identification reconfiguration (not (counter_delta = out)) No mode constraints mode = reset ⇒ (not (counter-delta = out)) Livingstone Server “ “likely likely” ” µ (t) Locked Locked Unlocked Unlocked o(t) ACE-TAO ORB (mode = unlocked) Λ (cmd-in = homeset) ⇒ homeset- homeset - Pseudo- Monitor cmd homeset- homeset - cmd likely” “likely “ ” reached reached- - s (t) cmd cmd Commands Values (next (mode = reset) homeset- homeset - cmd cmd Reset Reset ALLEGRO ORBLINK Monitors g g f f CLASH IPC Mode constraint: Telemetry Data Plant (not (counter_delta = out)) Chart: 7 Chart: 8 Chart: 9 Interferometer Model Delay Line Limit Switch Sensors Not all possible transitions shown • Inputs: none • Outputs: signal-out, health-state • Pair-Switches module consists of two sensors: one • Inputs: limit switch signals, home-sensor signal, soft limit switch, one hard limit switch counter-delta, DL-command • STB-3 soft limit switches are disabled • Outputs: servo-mode Chart: 10 Chart: 11 Chart: 12 Page 2
Executive: Executive: Demonstrated Scenario Autonomous Sequencing Autonomous Sequencing • robust, event-driven, goal-oriented, multi-threaded sequencing engine • Exec constructs: • Interactive commanding � general • written in rich procedural language (ESL): – (send-command :DL :back) � component-specific contingency handling – (send-command :DL :homeset) • Define “composite” (subsystem-level) constructs by layering and � Signaling failures, specifying recovery procedures • Nominal fringe acquisition scenario: combining component-specific constructs, e.g.: timekeeping – send Delay Line to front of track achievement & maintenance of fringe tracking � Integrating with external timekeeping sources (e.g.timeout specification) – calibrate PZT Dither goal achievement – homeset Delay Line (zeroes Laser Counter & locks IM) � Decoupling achievement conditions and achievement methods – slew Delay Line to estimated delay position – with Delay Line tracking, perform fringe search task management – Fringe Tracker locks onto fringe � Multi-threading, synchronizing concurrent tasks 1. activate & calibrate path dither • State achievement and maintenance (IM lock, DL tracking) logical database 2. establish internal metrology lock � Monitoring state variables of system 3. establish delay line tracking • Break IM lock, see autonomous recovery property locks 4. close fringe tracker servo loop � Coordinating concurrent tasks, controlling inter-task conflicts Chart: 13 Chart: 14 Chart: 15 Lessons Learned Future Work • Interface complexity • Continue MIR modeling and Exec sequence development – Livingstone/EXEC/RTC – importance of clear i/f specifications • Deployment to other testbeds • Scoping of models – primitive components vs. subsystems • Replace CLASH layer with a CORBA interface – “abstract” components Backup slides • Flexibility of IDL • Integrate Burton multi-step reactive planner – fixed RTC commands, telemetry – feedback from RA model development • Integrate C++ MIR system • System observability – inaccessibility of some desired telemetry • Replace Exec: • Multi-step recoveries � Model-based executive (MIT) – Livingstone implementation limited to 1-step reconfigurations � IDEA integrated planning/execution (Ames) • Commanded nominal transitions – nominal transitions conditioned on explicit commands – workaround via pseudo-commands Chart: 16 Chart: 17 Chart: 18 Page 3
SSI Sequence Interferometer Sequence Livingstone Algorithm up to 125 m Representative acquisition sequence • Initial state: Theory – L & R STs, L & R Sids, DL, FT idling 1. Move Spacecraft… (Including – LC not zeroed StarLight 2. Acquire 3. Acquire 4. Acquire Predicted State) right starlight angular metrology linear metrology Relay • Acquire starlight (L & R) Spacecraft – Slew Sid to estimated target angle (OL) Observations Diagnosis up to 600 m – w/ Sid tracking, perform ST search Truth Most Likely – ST locks onto target star Maintenance Best-first Candidate System Agenda • Acquire fringe Candidates 5. Acquire 6. Measure delay 7. Cool formation left starlight & delay rate – Calibrate PZT Dither Checked Solution – Homeset DL, zero LC, lock metrology Conflict- – Slew DL to estimated delay pos’n (OL) θ directed A* R – w/ DL tracking, perform fringe search Conflict Combiner Search – FT locks onto fringe Database All Conflicts New Spacecraft Conflicts • Perform science measurement 8. Find/track/measure fringe Lay & Dubovitsky – May 2000 Chart: 19 Chart: 20 Chart: 21 Path Dither Fringe Tracker SSI-specific Models Not all possible transitions shown Not all possible transitions shown • Inputs: PD-command, DL-command • Outputs: health-state • Inputs: FT-command Angular-metrology Siderostat • Model requires dither must be “on” before calibrating • Outputs: FT servo-state Chart: 22 Chart: 23 Chart: 24 Page 4
Exec Details Recovery Procedures 4. Dispatch Recovery Commands to RTC General Constructs Top Level Entry Points Initialize State Database EXEC Define Device Types 2. Recovery Define Raw CORBA Command Request 1. State Updates Define State Observation Predicates 4. Execute 3. Recovery Commands Define MIR/Exec CLASH-CORBA Command Suggestions Component Specific Constructs Livingstone Define Command Constructs for all Components Define Command Timeout Values • Livingstone implementation limited to single-step recovery suggestions • Multi-step recoveries had to be procedurally coded within Exec, e.g.: Define “to-achieve” Functions (defun RESET_DL (dl) Define Recovery Procedures ;;This is a recovery suggested by MIR to get metrology re-locked (achieve (:op_state :IM :locked)) (format t "Delay line recovery done (~s).” dl)) Chart: 25 Chart: 26 Page 5
Recommend
More recommend
