0 Planning in A] Agent World gyllabus i Blocks World, Component ofa Planing Sytem, Goal tack Plonning, Nonlinear Planning Constraint Posting, Hierarchical Planning, Reactive Systems, Other Planning Techniques. Contents rat 102 103 104 105 106 10.7 108 Blocks World. . Summor- 16 «+++ Planning. . sueeuetuavseesene, a, Goal Stack Planning Non-linear Planning using Constraint Posting. Summer- 18 - Hierarchical Planning Reactive Systems Other Planning Techniques University Questions with Answers (10-1)i atpny mu locks World Sm, t Inorder to compare the v: jety of met of pl ‘we should find tok a a of em inva age done tb coor croak tt be ile Soames canbe found PINE 7 SEE COHEN OME Or ty * There is fat surface on which blocks canbe placed. + There are a numberof quae Blocks ll of he sme ze * They canbe stacked one upon the oer: * “There i robot arm that ean manipulate the blocks Why Use the Bloca word ay an exsmpe? The blocks word is chosen because + 1s sulin simple and well behaved + Its easly understood «Yet stil provides a good sample environment oeudy planning 6 Problems canbe broken ino ney distin! sbproblems © We ean show how pra slo ned tobe combined to frm a reat complete sauton Actions of the robot arm ‘+ UNSTACK(A,B) * STACK(A,B) + PICKUP(A) * PUTDOWN(A) + Notice thatthe robot arm can hold only one block ata time. Predicates Inorder to specify both the conditions under which an operation may be performed and the results of performing it, we need the following predicates + ONAB) ONTABLES(A) * CLEAR(A) TECHWCAL PUBLICATIONS? Anup eeu kr knoe |. ON(A,B, $0) * ONTABLEG, SO)*CLEAR(A, so) we execute UNSTACK(AB) in state 50, then the resulting state SI HOLDING(ASI)*CLEAR@SI). ‘To enable the complete situation to be described, we provide a set of rules called tame axioms, that describe components of the state that are not affected by each or. 7 ONTABLE(&s) > ONTABLE(, DOUNSTACK(x3)s) + DO is a function that species fora gren sate and a given action, the new state that els from the exciton ofthe acon. gobot «problem solving system (STRIPS) ‘¢ List of new predicates that the operator causes : ADD, DELETE + PRECONDITIONS list contain those predicts that must be ru forthe operator tobe applied STRIPS style operators for BLOCKs World + STACK(xy) +P: CLEARG)*HOLDING(s) + D: CLEAR(y)SHOLDING(s) + A: ARMEMPTY*ONixy) + ICKUPG) =P: CLEAR(x) * ONTABLE(X) SARMEMPTY * D: ONTABLE(x) * ARMEMPTY * A: HOLDING(x) rma: HBLEATENS po on10.5 rit oom oui pasically involved moving from an init —" ey nd mg fi aw pul ate Ce ‘Art ttgenee to Pring A simplo search te atte AL a nsTack AB) e oi tm x Algo" wae fe seach xem dooreston tes, Frame Problem. yo Naor ' pu problem int inet par sous ronson “peinary search backtracking can be hard so introduce reasoning to devise heuristics ago contol te backtsacking, ve fist major method considered the solution asa serch fom the iia state 10 8 Mate rough a state space. There are many ways of moving through this space by fr seperate and the A" aol described the best fst search Brough 2 graph ses “Tod is fine for simpler probs but for more realistic problems it advisable s meteblem decomposition. Here the problem is split into smaller sub problems and (Sar solutions are then sys. The danger inthis method oxrurs when certain ee feome arte and have to be dicarded. How mich ofthe partial ston can [iired and Row muich needs to be recomputed. The fre problem ~ deciding which things change and which do not - gave some pidnece on enabling us to decide on what says the same and on what changes a2 we fom state to state Ifthe problem concerned designing a robot to build a car then wermting the engine on the chasis would not afect the rear of the car now-a-days. The AO* algorithm enabled us to handle the solution of problems where the problem eld be aplit into distinct parts and then the patil solutions reassembled. However Meracted with one another. Most problems have some thought into the ordering ofthe steps for eample if fe robot has to move desk with objects on it from one room to another; or fo move = sda fom one room to another and the piano is near the doorway. The thought proses tered in recombining. paral solutions of such problems is known as planning, AY As point we sum into 8 ascussion about the role of the computer in the design of 3 ln 'ts to how we can solve a problem. It is extremely unlikely at this sage (ht § ‘eoputer will actually solve the problem unless it is 2 game and here some in ‘ith a person is needed. 2) puTpowNea) = 3 GNTABLEISPCLEARIA)CLEAR(@)ONTABLE(A) Fig. 10.1.4 EEE Planning Pry seme] ‘The methods which focus on ways of decomposing the’ original problem into appropriate subparts and on ways of recording and handling interactions among tr Subparts a5 they are deteed during the problemsnlving process are often ead Planning, Planing refers to the process of computing several steps of problenrsa procedure before executing any of them. ‘aa Wat does planning Invoie 7 Planing problems ae hard problems : They are crayon + Solutions involve many aspect at we have studied so far: O° Serch and pablo slving stg © Koowlege sepresiaton schemes. * Prem deumpatin «beng pol ino smaller ples and ig solve these fst We have seen that i is pole fo solve a problem pi roblem by considering the appropri form of lnowleigereprectaton snd eng algerns ol and also to use searching methods, 6 sein netey ts oa op tt hy know TECHUCAL PUBLEATIONS® hap ctr nontdge rns? an TECHNICAL PUBLICAPa OA Ao ag Mito nteigence ae SAY ng ton le wt est thod of sprucing the enon f a prem, Ih oe see Sy : as 8 simulation if we are conden the handing of GBEUES a 2 apy fost office there are no actual people and so we on HY 2 Tange of possi = wise in a trafic control problem there are no cars o ica NE UP a a jan orci nS ei ee sian no the best solution we can then implement i at he ra st " This approach assumes that ther is conti in the way of fe. It cannot budge rapid change of revolution. Hor can lus approach ctr for unexpected eves sch faulty component of a spurious happeing such two items tick tpether org breakdown in the path somewhere val such as ina camera ree. When 8 flo ome tnrecogntsable state i encountered it nt necessary to resto for mck of wha ag ve ones a eid removing the needles fron partly compleced feted sweater, The problem sin restarting Kom a dead end ang this will need some backtracking, This method of solution is pursued to reduce the lve land so to ensure successful handling we must introduce reasoning to help to cater for faulls, To assist in controlling backtracking inital state been successfully solved is still of complexity in the backtracking required 1 many methods go backward from the goal ton + Components of a planning system 4. Choose the best rule to apply next based on the best available heuristic {information ‘Apply the chosen rule to compute the new problem state that arises from i application. 3, Detect when a solution has been found Detect when an almost correct soltion has been found and employ speci techniques to make it totally correct Detect dead ends so that they can be abandoned and the systems effort drecied in more fruit directions 4. Choose the rules to apply ‘The most widely used technique for selecting an appropriate rules to apply is fst to isolate a set of differences between desired goal state and then to identify those rules that are relevant to reducing those differences. + soe ls, ava of ler eu ifrmation canbe exploited 0 ho among them. For example, if we wish to travel by car to visita friend, then 4+ The fist thing to do isto fill up the car with fuel. [EEE TECHNICAL PUBLICATIONS? An op tut fr monte to-7 aa pad nat ave 8 HEN WE ted pe 1 aes tb ade 1 ems apg is systems, 2 les i easy. Each rule sim es — ald result from its application, ply specified the problem oa ex systems, We Tt Be able to deal with rules on net at ye al yi 9 dese, ore 2neh fh ngs mae oh ate description. rmber of approaches 10S ask ave Ben wed uns boone (1969) ry ae al el gs prey he sli of Iwrdpsider the problem of having two blads A and B sacked on each other ryan We ay fave a le yh cd edn MAB So) 4 ONTABLEC®, 5)» CLEAR(A, Sp) we now wish to UNSTACKA, B). We exes the operation flows [CLEARIxs) » ONY, 81) > [HOLDING(x, DO(UNSTACK(«y))) » CLEAR(yDO(UNSTACK(43)5))] where x, y are any blocks, $ is any state and DOO 5 fom the given action. ‘he result of applying this to state Sp to give state S, then we get: HOLDING(A,S,) CLEAR(B,S))- There are a few problems with this approach : till on the table. This needs to be encoded into Is ofthe state not affected by the operator. specifies that a new state results Twfame problem In the above we know that Biss! {fre axioms that describe component uaiication problem . “AC we reslve the frame problem the seuling decipon my 9 sradegats De oe nocd wo encode tat a boc canaot be peed on 1 of ie 7 so should this attempt fal ? op tt or none TECHRUCAL PUBLICATIONS? - AAAA ntetigence 10-8 PROTA tog I we a llow failure things can get complex - do we allow for alot of unlikely eye The ramification problem After unstacki know that A ching block A, previously, how do we know that A is no longer Una location? Not nt ad ose aly what As no ape ¢ Problem) itis hard to specify exactly what does happen. 3 State Description Approach In this approach for each tion, a change st makes 10 the Sate 18 descigg everything else remains unchanged The main advantage of this method is single mechanism (hat Is resolton) can perform all operations that aze required for state description. The major dlsadvantge the number of axioms that are required becomes very large if problemstate descipicn ts complex STRIPS Approach (1971) STIPS proposed another approach ‘© Basically each operator has three lists of predicates associated with it: © a list of things that become TRUE called ADD. © a list of things that Become FALSE called DELETE. © a set of prerequisites that must be true before the operator can be applied, + Anything not in these lists is assumed to be unaffected by the operation. ‘+ This method initial implementation of STRIPS - has been extended to include other forms of reasoning/planning (¢g. Nonmonotonic methods, Goal Stack Planing and even Nonlinear planning ~ We will see later) Consider the following example in the Blocks World and the fundamental operations STACK = Requires the arm to be halding a Bock A and the other block B to be dest Afterwards the block A is on block B and the arm is empty and these ae tue ~ ADD; The arm isnot holding a block and block B is not clear; predicates tat are false DELETE; UNSTACK ~ Requires thatthe lock A i on block B; thatthe am is empty and that block A is clear. Afterwards block B is clear and the arm is holding block A - ADD; Tht arm is not empty and the block A isnot en Hlock B - DELETE: TEOWCAL PUBICATIONS® 4a op prim hy Mowe 10-9 oe ee ee sete __—__anarg nft get ir now greatly reduced the information that needs to be held. if a new grouse we 0 edo anew acne feu opaton. Ube te method we femove the stale indicator and use a database of predicates ‘0 Fost cues sate en pe et sae a8: ‘at gtE(8) * ON(AB) © CLEAR) NTA stack operation the new sate is se Apts) * CLEAR) * HOLDINGA)» CLEAR) ing Progress Pega souution can be detected if an devise a predicate that is true when the solution is found and is false Pie 6 | gequires a great deal of thought and require a prot _.ovtcting @ sotutlon owe tanning system has suceeded in finding a slution to problem when thas rer sequence of operators that wansforms the inital problem state into the goals «How wil it knove when this has been done ? 1 Fp simple problem-solving systems, this queston is easly answered by & straightforward match of the state descriptions. «one ofthe representative systems for planning systems i, predicate logic. Suppose thr os.a part of our goal, we have the predicate P(e). To see whether Pts) is casted income state, we ask whether we can prove Px) given the asertions that dscribe that state and the axioms that define the world model ‘.oseeting Dead Ends " i «Asa planning system is searching for 3 sequence of operators to solve a past froblen, it must be able to detect when iis exploring » path that can never lead toa solution. + The same reasoning mechanisms that sed for detecting a dead end. * the search ia reasoning forward from th inital state, it can prane any Path that leads to a state from which the goal state cannot be reached. * search process is reasoning backward from the goal state it can also termite are pat the iitial state cannot be reached or because lite progress is being made. In backward reasoning each geal i decomposed v9 rls ach ‘ray lead to a set of additonal subgoals. The subgoals. Each of these subgoals major catoape io that every subgoal can lead to multiple subgoals making » Frcblem harder than original one can be used to detect a solution can often be TEGHIIGAL PUBLICATIONS? «An up Ort for arcuPronning aa eet, Arta non eons: 10-10 Detecting false tals Is also necessary: + Eg. At search ~ if insuifclent progress & made then this tails aborted ig f a more hopeful one. = ‘* Sometimes it is clear that solving a problem one way Parts that are harder than the original state. © By moving back from the goal state to the intial state It 15 possible to dy nlc aed aah wall pat that involves a confit can be pruned out, means that there are more paths et has reduced th probe, + Reducing the number of possible available for those left Supposing that the computer teacher is ill at a school there are to pos alternatives : hho knows computing or ‘© Transfer a teacher from mathematics + Bring another one in, Possible Problems : + If the mathematics solved. «+ If there is no money left the second solution could be impossible. iy decomposable we can treat them as decomposable ant If the problems are neat! then patch them, how ? Consider the final state reached by treating the problen x decomposable atthe curent state and noting the differences between the goal state ad the cument state, and the goal state and the initial state and use appropriate Means Exd analysis techniques to move in the best direction. Better is to work back inthe pat Tending tothe curent state and see if there are options It may be that one eptonal pd ‘could lead to a solution whereas the existing route led to a conflict. Generally this freans that some conditions are changed prior to taking an optional path through te problem. “Another approach involves putting off decisions until one has to, leaving dedsion teacher is the only teacher of mathematics the problem is ng making until more information is available and other routes have been explored. Often Time echons need ot be ten as these nodes are never reached Repairing an Aimost Corect Solon e The Kinds of techniques we aze discussing are ofteh useful in solving nexiy decomposable problems. ‘+ One good way of solving such problems is to assume that they are completely decomposable, proced to solve the sub-problems separate check tat roblems separately, and then ‘when the sub-solutions are combined, they do in fact yield a solution to the original problem. TEOAIEA PBLEATONE™- hn pat none ee | stack Planning go So, te probe talver this methe ‘makes use of a single stack that contains both noe ee en aon set Tle se ers 9 buat dae he ea an THe Fe epertsdsted ws PRECNETON, AD, ol DELETE ss | tick planing meted tds prbers invelvng cone = Peis int ae The gael one ts Bini ae Sy generated by this method contains a sequence of operator for attaining the followed by a complete sequence forthe second goal etc. ding sep of prem scvng pes, th op gal on esac int each SUCeee well be pursued. ‘sequence of operators that satisfies iti found, that sequence is applied to rte description, yielding new description. goal that i then at the top of the stack is explored and an attempt is the st isfy it, starting from the situation that was produced as a result of «Next the rade 10 sal satisfying the first goal | This process continues until the goal stack is empty. 1 phen as one last check, the orignal goal is compared to the final state derived trom the application of the chosen operators. If any components of the goal are ot stsed in that state then those unsolved parts of the goal are reinserted onto the stack andthe process i resumed. Gal stack planning To start with goal stack is simply «On (C, AJAON (8, D)SONTABLE(A)SONTABLE(D) ONG AON. OF SiatevoNTABLED) TECHNICAL PUBLICATIONS® - Aa up ts fo inowe®Se Pavnirg nM hen We, unstack) Acta {etl ratzonen sos P94 pagy * This problem ye fies ' separate into four subproblems, one oF e3ch componeny | me oul = Noo ante Tho ote spobiens OSTABLEA) ad ONTABLD) Ay nolan aYeHOLDINCIO al state Smee ft cKica rack + Altemative 1: Goal stack: gis : Goal stack oNt6,D? ON, a) * BRIG AISONT® DIMOTAD + ONG.) «We ace that we CAN PO predicates onthe sack tl we reach HOLDINGIC) for . need to find sultable operator which WE gre operators that might make HOLDING(Q) true : PICKUP(C) and UNSTACK(C2). Without looking ahead, since we cannot tell which of these ators is appropriate, we create two branches of the search tre corresponding eating gal ach: * ON (C AYONG, D)OTAD + Altemative 2 : Goal stack: + ONG, D) + ON (CA) + ON (GAON@D;OTAD “Exploring Operators + Pursuing alterative 1, we check for operators that could cause ON(C.A) it yes Of the 4 operators, there is only one STACK. + STACKIGA) + ON@,D) + ON(GA}ON@,D/VOTAD i a oS e CCLEAR(A)'HOLDING (C)__CLEAR(A}HOLDING (©) Preconditions for STACK(C,A) should be satisfied, we must establish them as subgoals + CLEAR(A) STACK(C. A) STACK) + HOLDING fone) ONY _ = + CLEAR(A)HOLDINGC) | ON(CAVONG,DVOT AD__ON(GA)ON(.D)‘OT AD + STACKIGA) Table 1034 Goal stack + ON@.D) ve “ha et an ene sven sein ee? beter than unstacking it because i 6 ot + We can tell that picking up C( alt 1) is curently on anything, So to unstack it, we would fist have to stack it This ‘would be waste of effort. * But how could the program know that ? ‘+ If we pursue the alternative 1, the fop elem: a! ‘which is already satisfied, so we pop it off. CLEAR(C) s Popped off. + ON(GA)ONG,DIOTAD ‘Hee we exploit the Heuristic that if HOLDING is one of the several goals to be achieved at once it should be tacked las. Next we see if CLEAR(A) is true itis not. The only operator that could make it true is UNSTACK(G, A). This produces the goal stack : + ON@AD + CLEAR) (ON (@ A}*CLEAR(#)*ARMEMPTY TECHICAL PUBLCATIONS® An op at fr owe mth goal stack is ONTABLEO) at n also satisfied and is TESA PUBUEATIONS? An pt wnetzence 10-14 Paene9 OA Fee, * The remaining precondition of PICKUP(C) is ARMEMPTY which isnot gy ince HOLDING) i tro we aply the operator STACKIRD) Ti mle + CLEAR@) + HOLDING®) + CLEAR) HOLDING) + STACKa.D) + ONTABLEICIACLEAR(O*ARMEMPTY + PICKUP + CLEAR(A)SHOLDINGIC) + STACKICA) + on@D) + ON(GAMONG,D/OTAD Complete plan 1. UNSTACK(CA) 2. PUTDOWN(C) 3. PICKUP(A) 4, STACK(A,B) 5. UNSTACK(AB) 6. PUTDOWN(A) 7. PICKUP) 8. STACK(B,C) 9. PICKUP(A) 10. STAKC(A,B) [IZ Non-linear Planning using Constraint Posting [SUERTE + Ditficult problems cause goal interactions, + The operators used to solve one subproblem may interfere with the solution fo & previous subproblem. ‘+ Most problems require an interwined plan in which multiple subproblems af ‘worked on simultaneously. + Such a plan is called nonlinear plan because it is not composed of a lineit sequence of complete subplans. TECHNICAL PUBLICATION? An up uate ho sats 10-18 rainy Wt Agent Wed 1 ccasider the SUSSMAN ANOMALY + ar seh 9 on Ge pele be wad on aren + ys a nonlinear plan using heuristics such as : + Thos ® hieve ON(AB) cleatng block A puting block C on the table. 1 Fe TON(B.) by stacking block B on block C. 2 Allee ON(AS) by sacking block Aon block B. 5 Com posting has emerged as a central technique in recent planning systems conetiGEN and TWEAK) aint posting bulls up 2 plan by: cate enon * ying to order them and ‘bindings between variables inthe operators and actual blocks. eT i ideas for the fan consists of mo steps and by studying the goal stale Cao ‘generated. There is no order or detail at this stage. Gradually more meen if subsets of the steps are le sProduced and constraints about the order of 3 seal ntl ‘a completely ordered sequence is created. In this problem samen ton ith end conditions ON(A,B) and ON(B,C) which indicat suggests two steps Wi sea a ENSTACK giving the layout shown below where the operator is preceded by ron ae followed by pst radios OP BARD) CLEARC) *HOLDINGIA) *HOLDINGE) STACK(AB) STACK.) ‘ARMEMETY ‘ARMEMPTY Ona.) one.) CLEAR) LEAR HOLDING(A) HOLDING Nore «+ There is no order at this stage. + Unachieved preconditions are starred ("- + Both of the HOLDING preconditions are una in the initial state. * Delete postconditions are marked by ()- chieved since the arm holds nothing TECHNIGAL PUBLICATIONS® - An up tut for nowleda®Atitctt intetigonce — 10-16 PO Ag y ae thethods have introduced heuristics to achieve B03 oF preconay Planning method brought all these together under one formalin 2% Feces that introduc ‘are mentioned in bracken, following sant mtuced/sed the following heres tioned in bracketg as “Constraint Posting * The idea of constraint build up a plan by: in constraint posting is 0 pa plan by increme hhypothesizing operators, partial orderings between Operators, and binding variables within operators * At any given time in the problem-solving proces, we may AVE a set of wef ‘operators but perhaps no lear idea of how those operators should be ordegy with respect to each other. ‘+ A solution is @ partially ordered, partially instantiated set of operators to gener an actual plan, we convert the partial order into any of a number of total orders, *Constraint posting versus state space search State Space Search, = Moves in the space : Modify world state via operator = Model of time Depth of node in search space = Plan stored in Series of state transitions Constraint Posting Search ‘© Moves in the space : Add operators 1 Order operators = Bind variables 1 Or Othenwise constrain plan © Model of Time 1 Partially ordered set of operators Plan stored in : = Single node TECHNICAL PUBLICATIONS® - An up eat fr psig Pranning in AI Kgont Word tion esi new steps (GPS). tlon rom ratcasing 2 step t0 go before another step (Sussmans HACKER). scone pein 2 new step Detfen fo steps to revert a precondition (NOAH, NONLIN). sone Esabishmont sr sgnng a value toa variable fo ensue a precondion (WEAN. separation m preventing variables being assigned certain values (TWEA. ww lets look atthe effect of each heuristic. ‘We must ry and achieve the preconditions of the STACK operation above. We could ay pking up te respective blocks: cana cuzano) conse) ‘corsa crn “acre ara) nea arama Tara Samer? emer wowona 0.2) a stconditions of this set could negate the ‘At the moment there is no plan a5 the pos recondition of the first (STACK) plan so we must introduce order: «Ifthe eventual plan contains a PICKUP then STACK step then. be satisfied by some other steps. jucing constraints whenever step addition 1p should precede a corresponding ‘+ HOLDING preconditions would need to * Solve this by enforcing ordering by introdi is employed. + In this case we need to state that a PICKUP ste STACK step. That isto sey PICKUP(A) + STACK(A,B) PICKUP(B) « STACK(BO) TECHNICAL PUBLIGATIONS® - An up wus for knoweog®‘tic tetigence TMs gives four steps partly ordered and four unacheved contons © CLEAR(A) - Block A fs not clear in initial state * “CLEAR(®) - Although block B is clear in initial slate STACK(A) ostcondition CLEAR(B) might precede the sep with *CLEAR() precondition” * Two *ARMEMPTY - inital state makes ARMEMPTY but PICKUP sep ),, ARMEMPTY and could again precede this step. We can use the promotion heuristic to force one operator t0 precede another so yu, the postcondition of one operator STACK(AD) dows not negate the preconditay ‘CLEAR(B) of another operator FICKUP(B). Tis ordering i represented by PICKUP@) — STACK(A) We can’use promotion to achieve one of the ARMEMPTY preconditions: Making PICKUP(®) precede PICKUP(A) ensures that the arm fs empty and aly conditions for PICKUP) are met This is writen PICKUP(B) < PICKUP(A). Unfortunately a postcondition of the first operator is that the arm Becomes not ‘empty, so we need to use the declablering heuristic to achieve the preconditions ofthe second operator PICKUP(A). Declobbering + PICKUP(B) asserts ARMEMPTY. ‘© But if we insert a step between PICKUP(G) and PICKUP(A) to. reassert ARMEMPTY then we can achieve the precondition. # STACK(BO) can do this so we post another constraint : PICKUP(B) STACK(B,C) + PICKUP(A) We still need to achieve CLEAR(A) The appropriate operator is UNSTACK(xA) by step addition. This leads to the following set of conditions 10-18 PAPO HAA a Ome variable x can be Bound to the block C by the simple establishment heuristic since The ein the initial state. The preconditions CLEAR(C) and ARMEMPTY are 8 on. etACK(B0) and by PICKUPG) or PICKUP) however, ‘we must introduce three orderings by promotion to ensure the operator 5 Ack(CA) insta (GA) lattarrow STACKB.C) neta UNSTACKICA) lsftarrow PICKUP(A) UNSTACKICA) Jetarow PICKED) involves adding a step and ths cobbers one of the oe ion ‘preconditions CGg) viz ARMEMYTY, aways petal problem wih ths bes nM yever all is not Tost a8 there is an operator, PUTDOWN that has the requced Howtjon and given that the operator UNSTACK(CA) had generated the ston for it of HOLDING(C) we ean produce an extra operator successfully eo Feouomvc(C) PUTDOWNIC) HOLDING(C) (ONTABLE(C) ARMEMPTY ‘This operator declobbers the operator PICKUP(®) yielding the sequence UNSTACK(C,A) « PUTDOWN(C) « PICKUP(®) This yields the final sequence : *CLEAR(3) ‘ ce : Beal *ARMEMPTY PUTDOWN( —__ 3. PICKUP(B) UNSTACE(EA) isc eg one) SPICKUP(A) ” ~ARMEMETY 6 STACK(A.B) HOLDING(x) CLEAR(A) TECHNICAL PUBLICATIONS® ~ An up trust fr knowedoe TEOIWICAL PURLEATONS™ An op aA Mic! inttigence 10-20 Pranning iA thy s ‘finish this section by looking atthe formal form of the TWEAK algorityy 1 Inititize $ tobe the st of propositions in the gos! tte 2 Repeat 8. Remove some unachieved propesiion P from S. B. Achieve P by using ene of the heuristics. Review all the steps, including additional steps 10 find ll unahi preconditions, add these to $ the set of unachleved preconditions uni) gy set Sis empty. 3. Complete the plan by converting partial orders into a total order performing ay necessary instantiations, binding of the variables. "Modal Truth Criterion +A proposition P is necessary true in a state Sif and only if to condition hai there is a state T equal or necesaily previous to $ in which P is ne asserted and for every step C possibly before § and every proposition Q coudesignating with P which C denies, there is a step W necessarily betveen C fra S which aserts Ra propostion such that R and P co-designate whenever P and Q co-designate + Mosal truth criterion tells us when a proposition is true EDF Hierarchical Planning + In onter to solve hard problems, a problem solver may have to generate Jag plans ‘+ tis important to be able to eliminate some of the details of the problem until 2 solution that addresses the main issues is found. i + Then an attempt can be made to il the appropriate detail. + Early attempts to do this involved the use of macro operators, + But in this approach, no details were eliminated from actual descriptions ofthe operator. + As an example, suppose you want to vista friend in Europe b in Europe but you have & limited amount of cash to spend. Fst preference will be find the airfares, sine finding an affordable flight will be the most difficult part of the task. You should not wry about geting out of your driveway, planning aleport jeway, planning a route to the airport €& stl you are sure you hve alight '* The assignment of appropriate criticality value is crit this ‘hierarchical planning method. ¥ ea. + Those preconditions tht no operator can satisfy are cletly the most cial TECHNICAL PUBLICATIONS® - Anup tat ar Inowiedge tee Praning Al Agent Wot Ce le, solving a problem of moving robot, for applying an operator, sell, «TP THIROUGH DOOK, the precondition that there exist a door big enough for UTobot 10 get through is of high cicality since there is nothing we ean do Ame tf 1 nok Ere psTRIPS AU ypSTRIPS actully planned in a hierarchy of abstraction spaces, in each of which 1 ApsTEaitions at Tower level of abstraction were nord. psTRIPs approach is a follows : rat solve the problem completely, considering only preconditions whose Pcality value is the highest posible. “These values reflect the expected difficuly of satisfying the precondition. ‘To do this, do exactly what STRIPS did, but simply ignore the preconditions of ower than peak criticality ‘Once this is done, use the constructed plan asthe outline of a complete plan Sha consider preconditions at the nextlowest criticality evel Because this approach exploreé entire plans at one level of detail before it looks rie lower-level details of any one of them, it has been called length-first approach. [DE Reactive Systems 1 The idea of reactive systems is to avoid planning altogether, and instead use the heervable situation as a cue to which one can simply react ‘A reactive system must have access to a knowledge base of eserves what actions should be taken under what circumstances, ‘A reactive system is very different from the other kinds of planning systems we have discussed because it chooses actions one at atime, «+ Te does not anticipate and select an entire action sequence before it does the first thing, + Example is a thermostat. The job of the th constant, «Reactive systems are capable of surprisingly complex behaviours, + The main advantage reactive systems Have over traditional planners is that they operate robustly in domains that are dificult to model completely and accurately. + Reactive systems dispense with modeling altogether and base their actions diectiy ‘on their perception of the world. f some sort that ermostat is to keep the temperature TECHNICAL PUBLIGATIONS® - An up trst fr knowAta aigonce wis Planning Al Aang o-22 + Another Qhother advantage of reactive systems is tat they ae extremely responsi ' Avoid the combinatorial explosion involved in deliberative lanaing, * This makes them attractive for real time tasks suchas diving and walking Thermostat Bing 1s an example for reactive systems ‘+ Its job is to keep the temperature constant inside a room. : + One might imagine a solution to this problem that requires significant amouny Planning, taking into account how the extemal temperature rises and ae the day, how heat flows from room to room, and s0 forth tog ‘+ Real thermostat uses simple pair of situation-actic 1. If the temperature in the room is k degrees above the desired temperature, th, tum the airconditioner on 2. If the temperature in the room is k degrees below desired temperature, they ‘tum the airconditioner off Other Planning Techniqu «© Triangle tables [Fikes et al,1972]- It provides a way recording the goals that ex operator t= expected to satisfy as well a the goals that must be true for its comet ‘hecution. Tf some exceptional situation occurs during execution of the plan then a table is provided that stores the information which is required to patch the plan Metaplanning [Stofik,198t a) - It is technique for reasoning about the problen being solved, which also provides a way organizing the planning process. Macro-operators [Fikes and Nilsson, 1971) - It allows a planner to build new ‘operators which represent commonly used sequences of operators. Case-based planning [Hammend, 1996] - It makes use of already existing plan lop new plans there hy providing rensaility of oxeting plans. ion rules Pe 7 iyi upg ins eer ens 13, 103,104 and 10) 2. a rae ter son 182) nar ene ts een pi sing nd ling? (Refer section 10.2) . Espa het fg ih ypc sch th abe aap (Bee vec 102) Wien TWEAK end STRIPS. le scion 13) wan mtd ele sean 0) Esplin po 10-29 “ Pacing nl Agent Hold << ——_—— wins 09 Tri9 ra nts tet mainionSSC* 7 ext wna ling mb, ler veon 0 5 arch! planning with example. (Refer vecon 105) 1 en mt mt ten ae gn he errs a ey cr pie preantn (dle (D) anda 11 Pe ection 102) emir he fling eprsntion nfm Bawa ‘ON, D) * ONC, B® ONTABLE (0) ® ONTABLE (A. (ON(A, B) * ONC, D) * ONTABLE () 8 ONTABLE (0) ‘Show howe STRIPS woul sce this protien, Shi hese proses pote pin pes nd ft jsify bt ca ede? (eer section 102) goa lak planning. Sole the fllocngasng gal ack planning. Refer section 103) 2 Elsi e a => the follwing sing goal stack planing, Refer section 102 © a oe ef 4 soe a D a a 15. Consider the problem of pron ASSIGN (2,8, i, 00) Jeo ication 2 which previously contsined the sue O. ASSIGN (0. 1, 09) 'P: CONTAINS (1b, 9) * CONTAINS (x, Oo) D: CONTAINS (x, 00) Ps CONTAINS (a, 10) Assume that there i atleast one atonal register C este 1a) What should STRIPS do with this problem 2 7 1) Hos might you design a program to sole his problem 15. Wintea note on recive system. (Refer section 106) University Questions with Answers 101 Explain goa stack planning using suitable example. (Refer section 102) o _somppng the content of to registers A and B. Suppoe thatthe single 1S arable which avg the cals, which & stored in lation c 2 (Refer section 102) = TECHHICAL PUBLICATIONS® » An up tt for Know