Full article in PDF

12 Pages·2004·0.2 MB·English

RoboCup in Higher Education: A Preliminary Report
ElizabethSklar1,SimonParsons2,andPeterStone3
1 DepartmentofComputerScience,ColumbiaUniversity
1214AmsterdamAvenue,NewYork,NY10027,USA
[email protected]
2 DepartmentofComputerandInformationScience
BrooklynCollege,CityUniversityofNewYork
2900BedfordAvenue,Brooklyn,NY11210,USA
[email protected]
3 DepartmentofComputerSciences,TheUniversityofTexasatAustin
1UniversityStationC0500,Austin,TX78712-1188,USA
[email protected]
Abstract. Sinceteam-based projects have been proven to be an effective ped-
agogical tool, we have been using RoboCup challenges as the basis for class
projectsinundergraduatecourses.Thispaperunifiesseveralindependentefforts
inthisdirectionandpresentsearlyworkinthedevelopmentofsharedresources
andevaluation. Weoutlinethreecoursesanddescribetherelatedclassprojects
inordertomakethecontextofourinvestigationclearandtomakeitpossiblefor
otherstoreplicateorextendourwork,andcontributetothesharedresource.
1 Introduction
Educationalrobotics,theuseofroboticsasaformofhands-onlearningenvironment,
isbecomingincreasinglycommonasrobotkitsarebecomingmoreaccessibleandaf-
fordable[31].Creativeinstructorsarefindingwaystoteachsciencetopicsusingthese
technologies, organizing tournaments around the robots; the energy, enthusiasm and
motivationdisplayedbystudentsofallagesisunsurpassed.We havefoundRoboCup
–especiallySoccerSimulationandtheRoboCupJuniorchallenges–tobeparticularly
conducivetocollege-levelclassroomuse.Theabilitytodemonstratetheoreticalmodels
andcomplexalgorithmswitha hands-on,accessible mediumstrengthensthe learning
experience.
Inthispaper,we documentourexperiencesincorporatingRoboCupactivitiesinto
undergraduatecourseswiththeideaofunitingotherswhoaredoingthesame.Ouraim
istwo-fold:one,tocreatearepositoryforrelatedcurricularmaterials;andtwo,tobuild
acommoninstrumentanddatabaseforevaluatingtheRoboCuplearningenvironment.
While we have foundthe link betweenRoboCup andtraditionalcourseworkin Intro-
ductoryRobotics,ArtificialIntelligenceandMultiagentSystemstobeanaturalone,we
presumethatthisarisesoutofourfamiliaritywithRoboCupthroughlongterminvolve-
mentwiththeinitiative.Indevelopingourrepository,wearehopingtomakethenotion
of incorporatingRoboCup into such coursework a relatively easy task for uninitiated
instructors,byprovidingsyllabi,readinglistsandprojectdescriptions.
D.Polanietal.(Eds.):RoboCup2003,LNAI3020,pp.296–307,2004.
(cid:1)c Springer-VerlagBerlinHeidelberg2004
RoboCupinHigherEducation:APreliminaryReport 297
Others have experimented with the RoboCup paradigm in undergraduate class-
rooms. Coradeschiand Malec used the RoboCup soccer simulator in a course on Ar-
tificial Intelligence Programming1 [11]. Birk has developeda course on Autonomous
Systemsthatusesthe Small-SizeRoboCupLeagueforpracticalexercises2.Vidaland
Buhler [34] have developed a series of graduate level courses on multiagent systems
usingtheRoboCupSimulationleague.
Wearealso,ofcourse,notthefirsttouserobotkitsinanundergraduateclassroom
as a hands-on learning environment. In 1989, Martin created the MIT Robot Design
project course (6.270) [16,18]. Yanco [36] has adopted this course, ending the term
withaBotball3tournament.Mataric’s“IntroductiontoRobotics”[20]takesahands-on
approach to the introduction of the basic concepts in the field of robotics and con-
cludes with a contest where robots play a ball game in a hexagonal field. There are
alsocoursesusinghands-onroboticsthatdonotfocusonteachingroboticsasthemain
subject. Littman’s“ProgrammingUnder Uncertainty”[15] teachesaboutmethodsfor
programmingunderuncertaintyandavarietyofmachinelearningtechniques.
Asidefromconstructingasharedrepositoryofcoursematerials,weareinterestedin
conductingacomprehensiveevaluationofthepedagogicalvalueofeducationalrobotics
in general and RoboCup activities in particular. Following on the work of Sklar et
al. [30], we are interested in trying to pinpoint the educational value of robotics and
the RoboCup initiative at the undergraduatelevel. Uniting multiple instructorsmeans
that we can not only share experiences, but also collect course evaluation data on a
granderscale,allowingustoperformanalysisacrossabroadercohort,witharangeof
academicaswellasculturalbackgrounds.
2 RoboticsinUndergraduate Education
Here,wedescribethreeclasseswherewehavesuccessfullyusedRoboCupchallenges
astermprojects.
2.1 IntroductiontoRobotics
Thisintroductorycourselooksatroboticsfromseveralaspects:technically,historically
and socially. Many of the technical aspects are based on Mataric’s course described
above. The course is designed for non-engineeringstudents to gain a hands-on expe-
rience with technology, as well as a basic understanding of the field of robotics and
thechallengesfacingthefieldtoday.Partofthecourseisspentreadinganddiscussing
classic material that relates to robots – including non-technical aspects such as sci-
encefiction,psychology,cognitivescienceandeducation.Theremainderofthecourse
takesahands-onapproachtointroducingthebasicconceptsinrobotics,focusingonau-
tonomousmobilerobots.LEGOMindstormsrobots4areused,andstudentsmustcom-
pletetwo projectswith them.First, theymustbuildrobotsto executea line-following
1http://www.ida.liu.se/˜silco/AIP/
2http://www.faculty.iu-bremen.de/birk/lectures/COURSES/
autosys.html
3http://www.botball.org
4http://www.legomindstorms.com
298 ElizabethSklar,SimonParsons,andPeterStone
(a)Line-followingmaze (b)Soccer
Fig.1.Robotcontests.
task culminating in a maze contest. Second, they construct robots to play soccer and
performinaRoboCupJuniorstyletwo-on-twotournament.
Thecoursebeginswithanintroductiontoroboticsandagent-basedartificialintel-
ligence[28].Thenwecoversomehistoryandthebasicsofbuildingandprogramming
withLEGOMindstorms[17,19],usingNotQuiteC5 [4,5].Therobotsareusedasex-
amplesfor the remainderof the topics, whichintroducethe generalareasin robotics:
effectors, sensors and control [18,23]. The area of control is covered in more depth,
discussing variousarchitecturesincluding deliberative,reactive, hybridand behavior-
based [1,6,8,22]. Learning is also discussed [13,21,35]. Other topics presented in-
cludeartificiallife[3,10],edutainment[14,32],cognitivescienceandpsychology[7,
24]andsciencefiction[2].
Thecourseistaughtovera14-weeksemester.Thereisone75-minutelectureand
one 75-minute lab per week. There are two exams, and students submit written lab
reportsdocumentingtheirsoftwareandhardwaredevelopments.Theyareencouraged
torecordresultsoftestsmadeandchangestotheirdesigns.Studentsalsoprepareaterm
project,presentedbothwrittenandorally.
SklartaughtthiscourseinSpring2001atBostonCollege6.Twenty-sevenstudents
were enrolled, three of whom were female. All were undergraduates, and there was
a mix of ages: first year (1 student), second year (3), third year (10) and fourth year
(13). The Computer Science Departmentat Boston College is in the School of Man-
agementandthereisnoengineeringschoolintheuniversity,sothehands-ontechnical
experienceofthesestudentswaslimited.SixteenmembersoftheclasswereComputer
Science or InformationTechnologymajors. The rest came from Biochemistry (1 stu-
dent),Communication(1),Economics(4),History(2),Marketing(1),Mathematics(1)
andPhysics(1).
The students were placed in groupsof three for workingon the roboticsprojects.
Since the experiencelevelsof the class was so diverse,Sklar assigned the groups,at-
temptingtobalanceeachgroupwithanequalnumberofbeginningandadvancedstu-
dents.Studentsweregivensomelabtimeduringthescheduledcourseperiodinorder
toworkontheprojects.However,thiswasnotenoughtimetoperfectrobotstoperform
wellin thecontests,so manyofthestudentsmetoutsideofclasstimeto workonthe
5http://www.baumfamily.org/nqc/index.html
6http://www.cs.columbia.edu/˜sklar/teaching/spring2001/mc375/
default.html
RoboCupinHigherEducation:APreliminaryReport 299
robots.Each studentwas requiredto submita lab reportindividually,which included
anassessmentofthecontributionoftheirteammates.Theeffortsofteammembersare
neverbalanced,howevertheinequitieswereobviousinreviewingthelabreports,even
withoutthepeerassessmentcomponent.Students’gradeswerebasedonthelabreports,
notontheirrobots’performanceinthecontests.
Thetwocontestswereheldinapublicspaceandstudentswereencouragedtoinvite
theirfriendstowatch.Theexcitementofthecrowdandthevisibilityoftheeventmoti-
vatedstudentstoworkharderafterthefirst(maze)contestinpreparingforthesecond
(soccer)contest.
The term projects presented a major challenge for these students, who were not
typicallyaskedtodoanywritinginComputerScienceclasses. Theywererequiredto
submit a brief project proposal several weeks prior to the final due date, in order to
getthemstartedandalsotoprovidefeedbackabouttheappropriatenatureofthetopic.
Therangeoftopicschosenwasquitebroad,fromtheuseofnanotechnologyinsurgical
robotstothehistoryofrobotsdatingbacktoancientGreece.Eachstudentgaveaten-
minuteoralpresentationontheirchosentopic.Thiswasdifficultformanystudentswho
werenotusedtospeakinginfrontofaclass.Althoughdiscussionfollowingthepresen-
tationswasencouraged,verylittleactuallyoccurredandseveralstudentsskippedclass
on presentationdayswhen they were notspeaking.Course evaluationresults (below)
confirmedthatthemotivationsurroundingthetermprojectwasminimal.
Students were given a survey at the end of the course. Forty-four percent of the
classresponded.Thesurveycollecteddemographicdataandalsoqueriedthestudents
about their learning experience. They were asked to identify which elements of the
coursewerehelpfulinlearningthematerialandwhichelementsoftheassessmentwere
valuable in helping them to solidify and demonstrate their knowledge of the subject.
TheresultsareshowninFigure2.
100 100
%)
helped learning (%)24680000 helped demonstrate knowledge (24680000 kccmrrrpppennitttyttdswmms: ======swmmmsooraaiccdizztcctteeeeeerrrnclmalocabronebentrpxeetroseaeprtsmpottortrt
0 labs notes lect read 0 cntmcntsmidrptmrptsrptwrpto fin rfipnto ==ofirnaallreexpaomrt
(a)elementsofthecoursewhichstudents (b)elementsofthecoursewhichstudents
identifiedashelpfultotheminlearning identifiedashelpfultothemindemonstrating
thematerial theirknowledgeofthematerial
Fig.2.SurveyResultsfromIntroductiontoRoboticscourse,Spring2001.
Overwhelmingly(83%),thestudentsfeltthatthelabs(i.e.,buildingandprogram-
mingtherobots)washelpfulforlearningthematerial,whereasonly33%saidthatthe
readingwashelpful.Seventy-fiveandsixty-sevenpercentrespondedthatthetwocon-
tests(mazeandsoccer,respectively)werevaluableinhelpingthemsolidifyanddemon-
strate their knowledge of the material. This confirms our intuition that the hands-on
componentsprovidemoreeffectivelearningexperiencesthanotheraspectsofcourse-
300 ElizabethSklar,SimonParsons,andPeterStone
work,particularlyattheintroductorylevel.Wespeculatethatthereadingschosenwere
perhapstooadvancedformostoftheclass.
Studentcommentswereoverallquitepositive,includingthefollowingstatements:
– “Greatcourse...lovedthelaxatmosphereandhands-onexperience.I’drecommend
thecoursetoanyCSmajor.”
– “Ithinktheclassideaisgreat.Itisagreathands-onexperiencetotryout.Thelabs
wereveryfuntimes.”
Therewere manycommentsthatthe mixedagegroupwashelpfulforall students,as
theinexperiencedstudentslearnedfromthemoreadvanced,andinassistingothers,the
advancedalsolearnedmorethemselves.Negativeremarkscenteredaroundrequestsfor
morelabtimeandlesstimespentonoralpresentations.
2.2 ArtificialIntelligence
ThemodernviewofArtificialIntelligence(AI)[28]isthatitisthestudyofintelligent
agents–autonomouscomputingsystemsthatperceivetheirenvironmentsandactupon
theminawaythatbothrespondstochangesintheirenvironmentsandworkstowards
underlyinggoals. Robots are prototypicalagentsthat have to movearound,and react
to,theirenvironmentsinpursuitoftheirgoals.Thusitishighlyappropriatetoexplore
areasofatypicalartificialintelligencesyllabususingroboticsprojects.
This course is designed to give a broad understanding of the basic techniques in
use todayforbuildingintelligentcomputersystems. Thesyllabusbroadlyfollowsthe
outlineof Nilsson’s ArtificialIntelligence:A newsynthesis[25].Studentslearn about
state-spacerepresentations,problemreduction,means-endanalysis,andreinforcement
learning. They study search methods including depth-first, breadth-first and best-first
search,aswellashill-climbingandalpha-betapruning.Predicatecalculusisintroduced,
alongwith variousmethodsoftheoremproving.Thecourse istaughtovera 14-week
semester,withtwo75-minutelecturesaweek,twoexamsandtworoboticsprojects.
ParsonstaughtthecourseforthefirsttimeatColumbiaUniversityinSpring20027.
Thirty-fivestudentswereenrolled,ofwhom6werefemale.Therewasawiderangeof
studentstakingthecourse–thebulkwereundergraduates(68%ofthe19forwhomwe
havethisdata),buttherewerealso6graduatestudents,bothMaster’sstudentsandPhD
students,andevenwithintheundergraduatestudentstherewasamixofagesfromfirst
year(1student),secondyear(1),thirdyear(8)andfourthyear(4).Themajorityofthe
undergraduateswereComputerScience majors(74%).Therestwere frombiology(1
student),economics(1),electricalengineering(1)andmechanicalengineering(2).
The students formed themselves into groups of three to four and had to program
LEGO robots in the Not Quite C programming language to perform two tasks – a
RoboCupJunior style line-following rescue task (in which the robot had to follow a
convolutedline,detectanobstacleandback-up,climbanddescendagradient,andfi-
nally detect and head towards a light source) and to play a simplified version of the
RoboCupJuniorsoccertask(therobotstartedatoneendofastandardRoboCupJunior
7http://www.cs.columbia.edu/˜sp/4701-2.html
RoboCupinHigherEducation:APreliminaryReport 301
two-on-twosoccer pitch, with the ball at the halfway point,and the robothad to ma-
noeuvretheballintotheoppositegoal,ratherlikea penaltykickintoanemptygoal).
Theculminationoftheprojectwasacontestinwhichthebasisforcompetitionwasthe
cumulativetime takentocompletebothtasks, andthestudentsalsowrote a reporton
theproject.Studentswerealsogiventheoptionofanextra-creditprojectofbuildinga
dancingrobotexactlyasintheRoboCupJuniordancecompetition.
Twocourseevaluationswereadministered.Onewasanofficialevaluationdoneby
the engineeringschool. The other was an informalpaper-and-pencilsurveygiven out
inclass.Theresultsoftheengineeringschool’sevaluationshowedthat55%ofthe33
studentswhorespondedgavetheroboticsprojecttheyundertookaratingof5(ona5-
pointscale)forinterest,andtwo-thirdsgaveitaratingof4or5.Twenty-onepercentof
thesamecohortofstudentsgavetheprojectaratingof5fortheamountlearnedduring
theproject,and58%ratedit4or5.
Theinformalsurveyprobedmoreintothestudents’perceptionofthevalueofthe
projectasopposedtootheraspectsofthecourse.Inparticular,studentswereaskedto
identifywhichaspectsofthecoursemostcontributedtohelpingthemlearnthematerial,
andwhichaspectsweremosthelpfultothemindemonstratingknowledgeofthemate-
rial.TheresultsaregiveninFigure3.Theseshowthatthestudentsfeltthattheproject
workwasnotashelpfulinlearningassomeofthemoretraditionalaspectsofComputer
Sciencecourses,butwasmoreusefulthanthetextbookandadditionalreadingmaterial
(whichnostudentsfeltwereuseful).Thepictureismuchthesameforthedemonstra-
tion of knowledge,with studentsratingthe contestas morehelpfulthan the final, but
less helpfulthan the homeworkand midterm. The reportwas rated least useful of all
(they really hated having to write a report). Despite the rather unencouragingfigures
fromthissecondsurvey,theveryobviousenjoymentthatthe majorityofthe students
tookintheprojectsencouragedustorepeattheexperimentthefollowingsemester.
100 100
%)
helped learning (%)24680000 helped demonstrate knowledge (24680000 khcmrfipenwnitytd:=====rhpmfioornibomdaojlteeteecwrcxtmooarnemrekptxeosartmt
0 hw proj notes lect text read 0 hw cnt mid rpt fin
(a)elementsofthecoursewhichstudents (b)elementsofthecoursewhichstudents
identifiedashelpfultotheminlearning identifiedashelpfultothemindemonstrating
thematerial theirknowledgeofthematerial
Fig.3.SurveyResultsfromArtificialIntelligence,Spring2002.
ThesecondofferingofthecoursewasgivenbyParsonsatBrooklynCollege,City
Universityof New York(CUNY) in Fall 20028. With the exceptionofthe absenceof
MastersandPhDstudents,thecohortwasbroadlysimilartothatatColumbiainterms
ofthefactorswemeasured.Eighteenstudentswereenrolled,ofwhom7werefemale.
8http://www.sci.brooklyn.cuny.edu/˜parsons/courses/
cis32-fall-2002/
302 ElizabethSklar,SimonParsons,andPeterStone
Again there was a wide range of students taking the course – all were undergraduate
studentsbuttherewasaconsiderablemixofageswith2secondyearstudents,3third
year,10fourthyearand2fifthyearstudents9.Themajorityoftheundergraduateswere
Computer Science majors (16 students). The remaining students were from Political
ScienceandBusiness.
In this offering,the projectwas givenmuchlater (dueto problemswith access to
the robots) and while the teams were the same size (3 students), the project took the
formofjusttheline-followingexercisedescribedabove.Againtheprojectendedwith
acontestandtherewasanextra-creditoptiontobuildadancingrobot.
Once again, an unofficial survey was administered, and the results are presented
in Figure 4. These results are a little more encouraging than those from Columbia.
Thistimetheprojectwasstillfelttobelesshelpfulinlearningthanlecturesorlecture
notes, but on a par with homework and more helpful than additional readings or the
textbook.In terms of demonstratingknowledge,the students felt that the project was
more helpful than either midterm or final. While gratifying, these figures should be
viewed with some suspicion. First of all, these students had self-selected to do robot
projects(studentswereallowedtodoanon-roboticsprojectinsteadiftheypreferred).
Second,asaresultoftheotherproject,thesefiguresarebasedonaverysmallsampleof
just11students.Finally,itseemsthatsomeofthereactionisbecauseitissounusualfor
studentsatBrooklynCollegetogettodoprojectwork(intheopencommentpartofthe
surveyseveralconfessedthatthiswastheonlyprojecttheyhadeverdone).Theeffect
ofthisinfluenceissupportedbythefactthatbroadlysimilarresultsweregeneratedby
studentswhodidthenon-roboticsproject(thoughtheverysmallnumberofstudentsin
thiscategorymakestheresultsextremelyunreliableandsotheyarenotpresentedhere).
100 100
%)
helped learning (%)24680000 helped demonstrate knowledge (24680000 khcmrfipenwnitytd:=====rhpmfioornibomdaojlteeteecwrcxtmooarnemrekptxeosartmt
0 hw proj notes lect text read 0 hw cnt mid rpt fin
(a)elementsofthecoursewhichstudents (b)elementsofthecoursewhichstudents
identifiedashelpfultotheminlearning identifiedashelpfultothemindemonstrating
thematerial theirknowledgeofthematerial
Fig.4.SurveyResultsfromArtificialIntelligence,Fall2002.
Commentsfromthestudentsurveysfrombothofferingsofthecourseinclude:
– “Whenworkingwiththerobot,Ilearntthatnothingisperfectintherealworld.A
lotoftimestheoutcomeisveryunexpected.”
– “ItremindedmeofwhyIwanttostayawayfromhardwareasmuchaspossible.”
– “Ithelpedimmensely!Ithelpedmeunderstandsomeoftheconceptscoveredinthe
lecture.”
9ItiscommonforCUNYstudentstotakemorethanfouryearstocompletesincemanystudy
part-time.
RoboCupinHigherEducation:APreliminaryReport 303
2.3 AutonomousMultiagentSystems
Autonomous agents and multiagent systems (AAMAS) is one of the fields to which
RoboCup participants have contributed consistently and prominently over the years.
DespitebeingthebasisforalargesubfieldofAI,thereisnogenerallyaccepteddefini-
tionofartificialintelligenceagents.Inlooseterms, agentsareprogramsthat(i)sense
theirenvironment,(ii)makedecisionsabouthowtoactbasedonthesesensations,and
(iii)thenexecutetheseactions.Autonomousagentsdoallthreeofthesestepsontheir
own,i.e.,withoutahumanintheloop.Multiagentsystemsarecollectionsofmultiple
agentsthatinteractwithoneanother.ThefieldofAAMAScoversawidevarietyofre-
searchfociandapplications,includingsoftware-basedinformationprocessing,robotic
controlofmultipleagents,entertainmentagentsandtutoringagents[12].
This course provides a broad introduction to autonomous agents with an empha-
sisonmultiagentsystems.Topicsincludeagentarchitectures,inter-agentcommunica-
tion,agentteamwork,distributedrationaldecisionmaking,agentmodeling,multiagent
learning and entertainmentagents. In additionto teaching aboutAAMAS, the course
aims to introduce undergraduates to the full spectrum of research activities engaged
in by professionalcomputer science researchers, emphasizing the differencebetween
these activitiesandtheactivitiesofa typicalundergraduatestudent[33].Assuch,the
courseincludesanopen-endedprogrammingproject,readingsfromtheresearchliter-
ature,publicspeakingandwritingrequirements,andopportunitiestocollaboratewith
peers. In order for students to succeed, they need to attain a mastery of the AAMAS
subject. However assessment is based primarily on their ability to engage in the full
rangeofactivitiesrequiredofresearchers.
Thecentralfocusofthecourseisasemester-longbuild-uptowardsaclassrobotic
soccercompetitionintheRoboCupSoccerServer[26].Studentsareassignedaseriesof
fourpreliminaryprogrammingassignmentsdesignedtogetthemfamiliarwithSoccer
Server and the CMUnited client code [27]. By the end of these preliminary assign-
ments, theyhavecreateda fullyfunctionalteam (althoughnotonethatisparticularly
competent).Thestudentsarethenencouragedtoproposeanimprovementonthisteam
asthetopicfortheirfinalprojects10.Forexample,onestudentproposedtousemachine
learningtechniquestotrainagoodgoaltenderwithoutpayingattentiontotherestofthe
team.
Themajorityofthereadingsforthiscourseareprimarysourceschosenbothtoin-
troduce particular topics and to engender some controversy(e.g., reactive [9]) versus
deliberative[29]agentarchitectures).Toencouragethestudentstocompletetheread-
ingsinatimelyfashion,theyarerequiredtosubmitabriefwrittenanswertoa single
questionpertainingtothereadingsatleast2hoursbeforetheclassstarts.Thefactthat
theresponsesareduetwohoursbeforeclassallowstheinstructortoincorporatethem
intotheclassdiscussion.Anothereffectofthequestionsisthatthestudentsdocometo
classpreparedtodiscussthereadings.Asaresulttherehavebeenmanyextendedand
heatedclassdiscussions.
An important componentof class participation is that each student is required to
moderateatleastoneclassdiscussionpertainingtothatweek’sreadings.Theyarein-
10Theyarealsogiventheoptiontoproposeaprogramming projectinamultiagentdomainof
theirchoice,buttypicallyfewstudentschoosetodoso.
304 ElizabethSklar,SimonParsons,andPeterStone
structedtoeitherdefendacontroversialstatementorposeaquestionandbepreparedto
defendeithersidedependingonhowtheclassreacts.Thisactivityturnsouttobeoneof
themostdifficultforthestudentstocomplete.Manyofthemarenotusedtospeakingin
frontofaclass,andtheyhaverarelybeenputinthepositionoffacilitatingdiscussions
asopposedtodefendingspecificpositions.
Thecourserequiresagooddealofwritingfromthestudents.Asabove,thestudents
are requiredto provideweekly written responsesto questionsrelated to the readings.
Theyreceivefeedbackpertainingtotheclarityandsoundnessoftheirresponses.More
significantly,thestudentsarerequiredtowritethreedocumentspertainingtotheirfinal
projects.First,theywriteprojectproposalsdefiningtheirgoalsfortheirprojectsaswell
as the proposed methods for achieving them. Second, they revise their proposalsand
add a section on their work in progress to create progressreports. Finally, they write
finalreportsintheformatofconferencepapers.
Studentsoptionallyworkinpairsonthefinalproject.Teamsmustwritetheirpro-
posalsandreportsindividually,withclearindicationsofwhatroleeachpersonplayed
in the collaboration;and as such, moreis expectedfromthem as a final product.The
robotic soccer project lends itself to such collaboration nicely since there are many
differentwaysinwhichthestudentscandivideupthework.
Theclassculminatesinasimulatedsoccertournament.Thestudentsaretoldatthe
outsetthatperformanceinthetournamentwillhavenonegativeimpactontheirgrades
(whileastrongperformancecanhaveapositiveimpact).Nonetheless,thetournament
isastrongmotivationalfactorforthestudents.Visitorsareinvitedtotheeventandthe
studentspresenttheirapproachesorallyandfieldquestionsastheirteamsareplaying.
Theperformancespreadamongtheteamsisoftenverylarge,especiallygiventhefact
thatsomestudentsdonotfocusoncreatingwinningteams.Allclass championshave
beentestedagainstamid-rangeRoboCupentryandlostsignificantly:despitestarting
with a fairly detailed client code base, the students are not able to attain competitive
world-classlevels.However,giventheirtimelimitationsthisfactisneithersurprising
nordiscouraging.
Stonehastaughtthiscoursetwice,firstatNewYorkUniversityintheFallof2001.
Fifteenstudentswereenrolled,onlyoneofwhomwasfemale.AllstudentswereCom-
puterSciencemajors.Allweregraduatestudents:12mastersand3Ph.D.Thesecond
offeringof the coursewas at the Universityof Texasat Austin duringFall of 200211.
Again,fifteenstudentswereenrolled,howevernonewerefemale.Thistime,thecohort
were undergraduates.Fourteen were seniors (fourthyear) and one was a junior (third
year). Most students were Computer Science majors, with the remaindermajoring in
ComputerEngineering.
Courseevaluationsandsurveyswereadministeredattheconclusionofbothcourses.
AtNYU,thecoursewasrated4.5outofapossible5(highestrating).AtUTAustin,the
coursewasrated4.6outofapossible5.Studentcommentshavealsoprovidedevidence
thatthestudentsappreciatetheopportunitytobeexposedtothevariouscomponentsof
scientificresearch.Bothtimesthecoursehasbeenrun,atleastonestudenthasdescribed
the coursein graduateschoolapplicationsasa primarymotivationforgoingonto do
research.Inaddition,twostudentsfromtheFall2002offeringareactivelycontributing
11http://www.cs.utexas.edu/˜pstone/Courses/378fall02
RoboCupinHigherEducation:APreliminaryReport 305
totheUTAustinentryintheRoboCup2003competition.Inonecase,theresearchis
leadinguptothestudent’sseniorthesis.
An informalsurvey was administeredin the middle of the term to the UT Austin
cohort.Studentswereaskedto ratethe programmingassignmentona scale of1to 5.
Thirty-threepercentgave it the highest rating; 57% gave it the second highest, while
10%scoreditaverageandnostudentsenteredlowmarks.Studentsalsoratedtheread-
ingassignmentsonthesamescale.Twentypercentgavethehighestrating;60%gave
thesecondhighest,13%scoreditaverageand7%gavethelowestrating.
Someofthecommentsfromstudentshaveincluded:
– “The discussions we have in class are quite unique,I haven’thad such involving
discussionsinanyclassbefore;”
– “Formatoftheclassisperfect.I’vewaitedthroughthreeyearsofcollegeforaclass
likethis....Iliketheclasssomuchthatmyotherclassesnowdisappointme;”and
– “TheonlythingIdislikeabouttheclassisthatwearelimitedinourapplicationof
ourknowledge.OureducationinAIisdirectedatimplementingaRoboCupsoccer
agent.Ifeelthatifwewereabletoapplyourknowledgetootheraspects,wewould
gainanevenbetterunderstandingofartificialintelligence.”
– “Thesimulatorcodewaskindatrickytounderstand.”
3 Discussion
Thethreecoursesdescribedaboveofferaninterestingcomparison,notonlyintermsof
contentandpresentationbutalsoinregardtothecohortsofstudentsenrolled.Collec-
tively,thecourseshavebeenofferedfivetimesatfivedifferentuniversities,providing
a broad range of backgrounds, demographics and experience levels – from first-year
undergraduatesataprivate,non-engineeringcollegetofourth-yearundergraduatesata
largestateuniversity,andincludinggraduatestudentsfrombothapubliccityuniversity
andtwolargeprivateuniversities.Thusthepositivefeedbackacrosstheboardinregard
totheroboticsprojectsisanencouragingandsignificantfactor.
Commentsaboutthe readingmaterialswere typicallyless enthusiastic, as the fig-
ures presented in the previous section indicate. Stone’s course, where students were
required to respond to the reading in short written assignments and were then given
the opportunity to discuss the readings formally in class, fared better than the other
courses, where reading was assigned in a more traditional manner – without written
readingresponsesandprimarilythematerialwaspresentedbytheinstructorsinlectures
where questionswere encouraged(but infrequent)and discussion was not the central
theme.Comparatively,Stone’sclasseshadfewerstudents,somoreeffectivediscussion
was possible. Nonetheless,severalstudentsfromall the coursescommentedthatthey
wished there had been better connectionsbetween the readingsand the projectwork.
This type of feedback is valuable to us and our colleagues in improving the existing
coursesaswellasdesigningnewones.
Thechallengepresentedtostudentswhowererequiredtomakeoralpresentations
to the classes (in both Sklar’s and Stone’s courses) is also notable. No matter what
careerpathisultimatelytaken,studentsneedtoknowhowtocommunicatetheirideas.
The development of oral presentation skills is important and should be encouraged,
despite students’ dislike of this aspect of the courses. Perhaps more creative ways of
oralreportingcanbeincorporatedintoallthecourses.