Stanford Artificial tnteliigence Laboratory July 1973 Memo AIM-228 Computer Science Department Report No. STAN-CS-74-409 The First Ten Years of Artificial Intelligence Research at Stanford Edited by Lester Earnest ARTIFICIAL INTELLIGENCE PROJECT John McCarthy, Principal Investigator HEURISTIC PROGRAMMING PROJECT Edward Feigenbaum and Joshua Lederberg, Co-principal Investigators Sponsored by ADVANCED RESEARCH PROJECTS AGENCY ARPA Order No. 457 COMPUTER SCIENCE DEPARTMENT Stanford University July 1973 Stanford Artificial Intelligence Laboratory tYlerno AIM-228 Computer Science Department Report No. STAN-CS-74-409 The First Ten Years of Artificial Intelligence Research at Stanford Edited by - Lester Earnest ARTIFICIAL INTELLIGENCE PROJECT John McCarthy, Principal Investigator HEURISTIC PROGRAMMING PROJECT Edward Feigenbaum and Joshua Lederberg, CoAprincipal Investigators ABSTRACT UnivSetrasniftoyrd at fieldrse lated ianantrdet lilfiigceinaclie n research of years ten first The have yielded significant results in computer vision and control of manipulators, speech recognition, citsiru e,hgnim mnaorigtoartpneserp e,ryroeht lacitamehtam yroeht fo,noitatupmoc dna gniledom fo cina glraoc.ismeeshscecorp sihT troper sezirammus e hsttnemhsilpmoccdana s esdeiivhopraprgoilbi bni hcae hcraeser.aera This research runs supported by the Advanced Research Projects Agency of the Department of Defense under Contract SD-153. The views and conclusions contained in this document are those of the authors and should not be interpreted as necessarily representing the official policies, either expressed or implied, of the Advanced Research Projects Agency or the U. S. Government. CONTOETFNA TBSLE i Section . egaP tceS noi Page 1. INTRODUCTION 1 3. CGINTISMIMRAUREGHORP PROJECT 39 2. ARTIFICIAL INTELLIGENCE PROJECT 2 1 .y3rammu Sfo smiAdna Accomplishments 39 2.1 Robotics 3 1.1.2 noitalupinaM 3 2.3 tnerruCytivitcA 40 21.2 Vision 5 3.3 swei VdesserpxE ybsrehtO 2.2 laciteroehTseidutS 10 Concerning DENDRAL 4 1 2.2.1 Mathematical Theory of Computation 10 secidneppA 2.2.2 noitatneserpeRyroehT 15 3.22 lacitammarGecnerefnI 16 .A SSECCA- OT NOITATNEMUCOD74 3.2 citsirueH gnimmargorP 17 B. THESES 49 2.3.1 Theorem Proving 17 2.32 citamotuAmargorP .C MLIFSTROPER 53 noitareneG 19 2.3.3 Board Games 19 .D LANRETXE.SNOITACILBUP 55 4.3.2 cilobmySnoitatupmoC 20 . A.E .I MEMO ABSTRACTS 67 4.2. larutaNegaugnaL 22 1.4.2 hceepSnoitingoceR 22 24.2 Semantics 24 5.2 gnimmargorP segaugnaL 27 1.52 LISP 28 2.5.2LIAF 29 3.5.2LIAS 29 6.2 retupmoCseitilicaF 30 1.6.2 ylraEtnempoleveD 31 2.6.2erawdraH 32 3.6.2 erawtfoS 32 2.7 detaicossAstcejorP 35 1.7.2 rehgiH latneMsnoitcnuF 35 2.7.2 latigiDyhpargoloH 36 3.7.2 dnuoSsisehtnyS 37 4.7.2 sraM erutciPgnissecorP 38 - . I. 1 .1 NOITCUDORTNI departments affiliated themselves with the Project, but without direct financial support: auhsoJ Lederberg ,)sciteneG( John la ieccinfeigtirlAletnI si eh tlatnemirepxedna Chowning and Leland Smith (Music), and theoretical study of perceptual and ynohtnA nraeH.)scisyhP( intellectual processes using computers. stI etamitlu laog si ot dnatsred neusehtsessecorp By early 1968, there were just over 100 well enough to make a computer perceive, people on the Project, about half supported understand and act in ways now only by ARPA. Kenneth Colby and his group elbissop rof.snamuh joined the Project that year, with separate gnidnuf morf the National Institute of In the late 1950s John McCarthy and latneM .htlaeH rehtO seitivitcayltneuqesbus Marvin Minsky organized the Artificial received some support from the National ecnegilletni Project at M.I.T. That activity Science Foundation and the National and another at Carnegie Tech (now Aeronautics and Space Administration. no l)lyetMi-seriegveinnr UadCid hcum foeht Zohar Manna joined the Faculty and the gnire ehncoriapeser ni l.aeiccniefgiitlrlaetni ,tcejor Pgniunitnoc sih krow nilacitamehtam yroeht fo.noitatupmoc nI 1962, McCarthy came to Stanford University and initiated another A. I. Project In June 1973, the Artificial Intelligence here. He obtained financial support for a Laboratory (as it is now called) had 128 llam sytivitca 6( )snosrep morf ehtdecnavdA members, with about two-thirds at least hcraese RstcejorP ycnegA )APRA(gninnigeb partially ARPA-supported. Other Computer enuJ ,51.3691 ecneicS ytlucaF srebmem ohw evahdeviecer such support include Robert Floyd, Cordell A retup meoc CnteniecmStrape Dsawdemrof ,neerG dna dlanoD.htunK ta drofnatS ni yraunaJ .5691 yB tahtemit there were 15 people on the Project, ehT citsirueH lardneD tcejorP retal(degnahc including Edward Feigenbaum who had just ot citsirueH )gnimmargorP saw demrofni .devirra Shortly, a decision was made to 1965 under the leadership of Edward expand the seitivitca of the Project, muabnegieF dna auhsoJ .grebredeL tIsaw yllaicepse ni eht aera fo eye-dnah.hcraeser initially an element of the A. I. Project and Additional support was obtained from consisted of five or so people for several APRA dna a 6-PDP retupmoc metsyssaw .sraey ordered. Lester Earnest arrived in late 1965 to handle administrative responsibilities of The Heuristic Dendral Project became a the expanding Project. separate organizational entity with its own APRA tegdub ni yraunaJ 0791 dna osla By the summer of 1966, the Project had obtained some support from the Department nworgt ueolbaliava supmac ecaps dnadevom fo ,htlaeH ,noitacudE dna.erafleW It has to the D. C. Power Laboratory in the dah 51 ot s0r2eb m ne.timsnhetcneorm sllihtoof evoba- .drofnatS ehT wenretupmoc metsys saw dereviled .ereht ruhtrAleumaS The following sections summarize and Jerome Feldman arrived at about this accomplishments of the first 10 years (1963- time and D. Raj. Reddy joined the faculty, .)3791 secidneppA tsil lanretxe,snoitacilbup having completed his doctorate on speech ,seseht mlif ,stroper dna stcartsba fohcraeser noitingocer as a member of the Project. d esctur doyopbrepr r.ufofats Several faculty members morf other 2 I .2 ARTIFICIAL INTELLIGENCE citsirueHgnimmargorP PROJECT ruO troppus fo s’nraeH krow on symbolic computation led to the development of The work of the Stanford Artificial REDUCE, now being extended at the Intelligence Project sah been basic and ytisrevinU fo hatU dna ylediw used deilppa hcraeser ni laicifi tercanegilletnidna .erehwesle closely related fields, including computer vision, speech recognition, mathematical Work in heuristic programming resulting in theory of computation, and control of an Luckham’s resolution theorem prover. This laicifitra.mra icsu rrentlayb outthb ee stth eorepmr oveirn existence, and it puts us in a position to test tenthe over funds ARPA Efxrpoemn ditures the limitations of current ideas about sr.aey beginning June 15, 1963 have scitsirueh os ew nac og dnoyeb.meht detnuoma ot 2.9$ .noilli mtuobA %34 fosiht saw rof lennos r,espe‘iralas( ,segaw,)stifeneb noitatneserpeR Theory 26% for computer and peripheral equipment ,esahcrup (tnemecalper ,strap dna ,)latner%8 Work in the theory of how to represent for other operating expenses, and 23% for information in a computer is fundamental tceridni .stsoc rof citsirueh ,gnimmargorp rof language understanding by computers and rof Here is a short list of what we consider to smargorp taht nac nrael morf .ecneirepxe have been our main accomplishments. More drofna tsSah neeb eht redael ni siht.dleif complete discussions and bibliographies Mathematical Theory of Computation follow. Robotics Our work in mathematical theory of noitatupmoc si demia ta gnicalpergniggubed tnempoleveD fo noisiv smargorp rof,gnidnif yb computer checking of proofs that identifying and describing various kinds of programs meet rieht specifications. objects in three dimensional scenes. The ,yhtraCcM ,renliM ,annaM ,dyolF,ihsaragI sen eecdsu lsctnciejb ohtiw talf secaf dnaosla and mahkcuL have been among the leaders devruc.stcejbo in developing the relevant mathematical theory, and the laboratory has developed the The development of programs rof tsrif lautc agnikcehc-fo osrmpargorp tahtnac manipulation and assembly of objects from check proofs that actual programs meet their parts. The latest result is the complete .snoitacificeps In particular, Robin Milner’s ylbmessa fo n aelibom orteutaaw.pmup LCF is a practical proof checker for a revised version of Dana Scott’s logic of Speech Recognition elbatupmoc functions. Development of a system for recognition of hcraeseR seitilicaF continuous speech, later transferred to Carnegie-Mellon University and now being eW evah depoleved a yrotarobal htiwyrev dednapxe nopu elsewhere. good computer and program facilities and special instrumentation for the evoba,saera including a timesharing system with 64 enilno yalpsid.slanimret LAICIFITRA ECNEGILLETNITCEJORP 3 We developed a mechanical arm well suited 1.2scitoboR to manipulation research. It is being copied dna desu yb rehto.seirotarobal Thep rojecth asp roduceds everals ubstantial :stnemhsilpmocca na citamotuanoitalupinam eW dengised na tneiciffe yalpsiddraobyek metsys elbapac fo gnilbmessa a retaw,pmup that has been adopted at several tenAPRA a laser ranging apparatus, and programs facilities. We invented a video switch for hcihw mr o sfcnioliotbpmiyrsc sfeodxelpmoc ysamlep tstsiaydhst si g nyi.ledebediipwoc troffe rojam a desucof won evah eW .stcejbo no scitobor ni lai.rntosiutdanmiotua nI eht esruoc fo gnipoleve dr u,oseitilicafew evah devorpmi ,PSIL depoleved nadednetxe 1.1.2 lloitalupinaM loglA relipmoc dellac ,LIAS dna detaerca document compiler called PUB (used to In 1966, we acquired and interfaced a ecudorp siht.)troper prosthetic arm from 0hcnaR soL sogimA Hospital. Although it had major mechanical ruO d eytlnreaiero-ep yttxeelteT rotidedellac shortcomings, the software experience was SOS has become an industry standard and .elbaulav A margorp saw nettirw roftniop ruo wen yalp sriodtide ”E“ si hcum.retteb to point control of arm movements. Computer servoing was used from the We have written utility programs for the beginning and has proven much more PDP- 10 and made numerous improvements versatile than conventional analog servoing. ot gnirahs-emit .smetsys Many of our A simple system that visually located blocks ,smarg oyrlpralucitrap ,PSIL ,LIAS dna,SOS scattered on a table and sorted them into .er adesu n isnezo df .o rsreretehuttpnomeocc stacks according to size was operational by eht gnirps fo 7691 elgniP[.18691 We designed an advanced central processor that is about 10 times as fast as our PDP-10. In order to move through crowded In support of this work, we developed ,secapskrow a margorp saw nettirw otdiova interactive design programs for digital logic selcatsbo elihw gniyrrac tuo mrastnemevom that have since been adopted by other repeiP[ 19681. That program was fairly hcraeser dna lairtsudni.snoitazinagro general but rather slow, since it used a local and not very smart search technique to inch Trairring sti yaw dnuora.selcatsbo nI eht 3791-3 6,9d1oirep 72 srebmem foruo A hydraulic arm was designed and built ffats dehsilbup .D.hP seseht sa laicifitrA according to stringent criteria of speed and Intelligence Memos and a number of other strength. Kahn developed a minimum-time graduate students received direct or indirect servo scheme, which is close to a bang-bang .tcejorP .I .A eht morf troppus n h oaf vK7r[9 efs 1. ethecTvei fssfsaeewrpmi (even frightening) but hard on the The following subsections review principal equipment. activities, _ with secnerefer to published selcitra dna.stroper The next arm was designed with software in d nniammni e.h1c9S6[91 tI sa wdetelpmocni December 1970, and has proved a good hcraeser;rotalupinam several other groups evah edam.seipoc 4 LAICIFITRA ECNEGILLETNITCEJORP mrA lortnoc erawtfos rof eht wen mrasaw ,no idteadtinveoirrpo ew evah a ledom foeht tilps otni a llams lortnoc-mra ovres,margorp object. which ran in real time mode on our PDP-6 propgtlrraaanmjneicntgo rya ancdo mputer, Arm planning is based on making complete [Paul 197 ,31 written in a higher level motions, e.g. picking up an object and language and running on a timeshared putting it down. If we pick up an object -PDP 10. arbitrarily, without thought for putting it down later, we may not be able to put it The arm servo software contained several down as desired and may need to grasp it new features: a) feedforward from a again. Complete trajectories are pieced Newtonian dynamic model of the arm, rehtegot morf suoirav ,stnemges .g.e,psarg including gravity and inertial forces; b) departure, mid-segment, approach, release. kcabdeef sa a yllacitirc depmadcinomrah Trajectories are specified by the user in an rotalli cgsnoidulcn iytico lneovitamrofnimorf interpretive hand language (HAL) in terms ;sretemohcat )c yr ontocietjaacrityftiidloimcaf fo sorcam ta eht level fo gnitresni a.wercs in the servo program, which allows tniopdnE snoitisop dna snoitatneiro eranetfo considerable change morf planned specified by positioning the arm itself in the trajectories to domoccaeta contingency proper position and recording joint angles, a .snoitidnoc erapmoC siht lortnocmsinahcem mrof fo gninrael“ yb .”gniod ehTegaugnal with the usual analog servo; the kinematic se dsieviotriplica frof lortnoc gnisu hcuotdna model and computer servo allow changes of ecrof.gnisnes lareve ssredro fo edutingam ni ehttnelaviuqe servo constants, to account for variations in ylbmessA pmuP ytivarg sdaol dna laitre.nsitceffe ruO tsrif rojam ksat saw ylbmessa fo na The arm was designed so that solution for elibomo trueataw .pmup A mlif hcihwswohs joint angles from position and orientation is the process in detail is available for simple and rapid. The techniques apply to a distribution elgniP[ dna ,luaPdetamotuA“ wide range of arms; thus our work has Pump Assembly”]. We describe the task n.i significance for industrial and other robotics some detail to show the level fo .snoitacilppa programming. ehT ECNATSNI-EVOM ytilibapac luaP[ The pump parts include pump base, cover, 11791 swohs a elpmis mrof fognitamotua .swe rdtcnesaksag We chose a plausible some manipulation procedures. The routine industrial environment with tools in fixed chooses a best grasping and departure for a places, screws in a feeder, and pump body class of known objects. Models of these and cover on a pallet. It is located by vision, objects are stored in order to choose all then moved by the arm to a standard elbiss ogpnips asrngoitisop -- lellarap ,secafa position, up against some stops. Pins are ecaf dna a lellarap ,egde .cte ehT llufegnar inserted into screw holes in the pump body of possible departure and approach angles is ot ediug tnemngila fo teksag dna .revocfI investigated and a solution chosen, if yrasse ceehnt d nsaehhcrae sot taes eht.snip possible, which allows manipulation in a e l ,g,ennsioisiwtroemh ta Onoitulo ssinesohc The gasket is placed over the guide pins and which uses an intermediate position to yllausi vdetcepsni rof tcerro c.tnemecalpehT regrasp the object. Thus, this facility cover is expected to be within about a provides a method for grasping general quarter of an inch of its fixed place on the polyhedra in arbitrary position and pallet. After locating the cover by touch, the
Description: