CDC related research at the Dept. of Computer Science TUT Jüri Vain CDC Workshop - Tallinn, Jan. 21-22, 2008 Research lines model-based planning in... FMs for model checking using... model-based test generation in... single agent systems multi-agent systems (Timed) automata learning Proof techniques for swarm coordination algorithms State isomorphism based symmetry reduction state space reduction by abstraction Iterated search refinement with bit-state pruning determinitic systems Test trace generation using MC non-deterministic systems Synthesis of test goal directed reactive testers Research lines model-based planning in... FMs for model checking using... model-based test generation in... single agent systems multi-agent systems (Timed) automata learning Proof techniques for swarm coordination algorithms State isomorphism based symmetry reduction state space reduction by abstraction Iterated search refinement with bit-state pruning determinitic systems Test trace generation using MC non-deterministic systems Synthesis of test goal directed reactive testers State isomorphism based symmetry reduction Problem Exploit symmetries to reduce the effects of state space explosion in explicit state reachability State isomorphism based symmetry reduction Solution Given a model in terms of unordered data structures, like sets and maps, and objects: Construct the state graphs of the states at run time and Use the computation of graph isomorphism to determine whether a state with similar structure has been seen. Results developed with Margus Veanes and Colin Campell and published at FORTE 2007 (June 2007) and in the PhD thesis of J. Ernits (Nov. 2007) State isomorphism based symmetry reduction Applications: Method is implemented in model-testing toolkit called NModel (by Margus Veanes et al, MS Research Redmond) There is similar implementation developed independently by Arend Rensink in the GROOVE project Research lines model-based planning in... FMs for model checking using... model-based test generation in... single agent systems multi-agent systems (Timed) automata learning Proof techniques for swarm coordination algorithms State isomorphism based symmetry reduction state space reduction by abstraction Iterated search refinement with bit-state pruning determinitic systems Test trace generation using MC non-deterministic systems Synthesis of test goal directed reactive testers Iterated search refinement with bitstate pruning Problem Calculate reachability in models that Run out of resources without producing results with known methods Produce too long traces to the error/desired state with known methods Such reachability problems can be used for solving problems related to Scheduling Hardware synthesis Offline test generation Iterated search refinement with bitstate pruning Solution Combine two well known methods in a new way: The key idea is to use collisions in bitstate hash table to randomly filter the search space. Each iteration requires very small amounts of memory A lot of prefixes of paths are covered multiple times, but Provides good results in several examples. Results published in the PhD thesis of J. Ernits (Nov. 2007) Iterated Search Refinement, and Bitstate hashing Iterated search refinement with bitstate pruning Practical applications 2 well researched case studies: Synthesis of a memory arbiter for a radar memory case study (J. Ernits, 2005) Calculating offline test sequences for model-based testing (J. Ernits, A. Kull, K. Raiend, J. Vain, 2006) The method cannot be used for disproving the reachability, but provides a new alternative for finding a witness trace. Pre-set test trace generation using MC Case study: modified INRES protocol Pre-set test trace generation using MC Results Time spent for finding sequences for the 2-switch coverage criterion Experiments made with Uppaal Cora Research lines model-based planning in... FMs for model checking using... model-based test generation in... single agent systems multi-agent systems (Timed) automata learning Proof techniques for swarm coordination algorithms State isomorphism based symmetry reduction state space reduction by abstraction Iterated search refinement with bit-state pruning determinitic systems Test trace generation using MC non-deterministic systems Synthesis of test goal directed reactive testers Synthesis of test goal directed reactive testers Problem: Given a Non-deterministic EFSM MSUT of System Under Test testing goal in terms of a set Trp of MSUT transitions Find EFSM MTSR s.t.. MTSR is I/O compliant with MSUT any trace of MSUT || MTSR includes labels of all transitions in Trp MTSR chooses a transition from those labelled with current input from MSUT that maximizes some gain function G. Synthesis of test goal directed reactive testers Solution: Complexity: Algorithm of constructing MTSR For all controllable transitions of the MTSR the upper bound of the complexity of the computations of the gain functions is O(|ESUT|3). At runtime each choice by the tester takes O(|ESUT|2) arithmetic operations to evaluate the gain functions (ASE2007, Vain, Raiend, Kull, Ernits) Synthesis of test goal directed reactive testers Experimental results All Transition Test Purpose One Transition Test Purpose Research lines model based planning in... FMs for model checking using... model based test generation in... single agent systems multi-agent systems (Timed) automata learning Proof techniques for swarm coordination algorithms State isomorphism based symmetry reduction state space reduction by abstraction Iterated search refinement with bit-state pruning determinitic systems Test trace generation using MC non-deterministic systems Synthesis of test goal directed reactive testers Proof techniques for swarm coordination algorithms Problem: proving properties of swarm (distributed) coordination algorithms self-stabilization, covergence, ... can be reduced to reachability analysis reachability analysis collides with scalability barrier Proof techniques for swarm coordination algorithms (ongoing work) Some solutions: Pattern based problem encoding provides right level of abstraction (spec. patterns are problem oriented) General principles of proving: Compositional approach combining different techniques of component proofs – MC, deduction,... Global properties by structural induction on the size of swarm/task Base: prove sufficiency of assumptions on a swarm fragment of some tractable size, e.g., using MC with symmetry reduction, (typically the size of fragment can’t be trivial either) (see Int. WS on HAM, Vain, Kuusik, Tammet, 2006) Step: Show that the induction step does not violate the assume part for the new fragment defined by the step Proof techniques for swarm coordination algorithms Demo: dynamic cleaning problem different zones of the room deteriorate with different rate a swarm of cleaning robots should keep deterioration below treshold robots of the swarm communicate trough pherromone trace Problem: given a treshold CTresh and initial value of the deterioration vector Sdet(0), s.t. Sidet(0) > CTresh for all i, 1) 1) prove that the swarm is always able to reach the state Sdet(t), s.t. Sidet(t) < CTresh for all i, and for all t’ > t, condition Stdet (t’ ) < CTresh is invariantly true. Proof techniques for swarm coordination algorithms Demo Research lines model based planning in... FMs for model checking using ... model based test generation in ... single agent systems multi-agent systems (Timed) automata learning Proof techniques for swarm coordination algorithms State isomorphism based symmetry reduction state space reduction by abstraction Iterated search refinement with bit-state pruning determinitic systems Test trace generation using MC non-deterministic systems Synthesis of test goal directed reactive testers (Timed) automata learning Problem: Planner synthesis for a human adaptive Scrub Nurse Robot (SNR) human adaptive = on-line learning from H-H/H-R interactions learning = constructing/modifying a human motion model (Timed) automata learning Context: SNR control architecture Deliberative ctrl. layer Reactive action planning NN model of Surgeon’s motions Motion recognition SNR’s “world” model Scenario of the surgery Predicted Recognized further motions motion SNR action to be taken Targeting and motion control Hand position data samplings Reactive ctrl. layer Surgeon’s behaviour model Nurse’s behaviour model Control Fource & position parameters feedback Collision avoidance 3D position tracking Direct manipulator control (Timed) automata learning Learning algorithm Input: Time-stamped sequences of observations of phase switching (set of timed traces TrT(Obs)) Parameters for state rescaling operator - distinquishing model observables/controllables - TA defining quotient state space Output: Extended (Uppaal version) timed automaton TA s.t. TrT(TA) = (TrT(Obs)| ) /~ % bisimilarity of timed traces TA (see also Vain, Miyawaki, MIC 2008, Insbruck) (Timed) automata learning (example) Time Chest X Elbow X Chest Y Elbow Y Chest Z Elbow Z Action 1 153 60 158 189 342 151 idle 17 126 127 147 255 356 150 prepare_instr 42 13 562 126 464 341 156 pick_instr 48 27 551 135 465 340 159 hold_wait 70 50.3 402 174 455 338 171 pass 78 152 115 171 258 352 164 withraw 88 153 60.2 158 189 342 151 idle 109 147 84 163 219 360 169 stretch 122 126 109 157 253 359 164 wait_return 139 96 159 151 286 358 157 receive 146 59 591 134 444 367 158 move_back 152 35 549 155 461 367 158 put_on_tray 156 153 60 158 189 342 151 idle 175 14 561 127 463 340 157 pick_instr 182 27 550 134 467 343 161 hold_wait 196 50 401 172 457 337 171 pass 205 128 110 158 253 357 161 wait_return 225 95 160 149 287 356 155 receive 233 56 589 133 445 366 159 move_back 239 34 548 156 460 369 157 put_on_tray 243 153 60 157 189 342 150 idle Thank You!
© Copyright 2026 Paperzz