Humanoid Robots: Modelling and Control

498 Pages·2018·29.46 MB·English

by  Dragomir N. Nenchev, Atsushi Konno, Teppei Tsujita

HUMANOID ROBOTS
HUMANOID
ROBOTS
Modeling and Control
Dragomir N. Nenchev
TokyoCityUniversity
Tokyo,Japan
Atsushi Konno
HokkaidoUniversity
Sapporo,Japan
withcontributionby
Teppei Tsujita
NationalDefenseAcademy
Yokosuka,Japan
Butterworth-HeinemannisanimprintofElsevier
TheBoulevard,LangfordLane,Kidlington,OxfordOX51GB,UnitedKingdom
50HampshireStreet,5thFloor,Cambridge,MA02139,UnitedStates
Copyright©2019ElsevierInc.Allrightsreserved.
Nopartofthispublicationmaybereproducedortransmittedinanyformorbyanymeans,electronicormechanical,
includingphotocopying,recording,oranyinformationstorageandretrievalsystem,withoutpermissioninwritingfrom
thepublisher.Detailsonhowtoseekpermission,furtherinformationaboutthePublisher’spermissionspoliciesandour
arrangementswithorganizationssuchastheCopyrightClearanceCenterandtheCopyrightLicensingAgency,canbe
foundatourwebsite:www.elsevier.com/permissions.
ThisbookandtheindividualcontributionscontainedinitareprotectedundercopyrightbythePublisher(otherthanas
maybenotedherein).
Notices
Knowledgeandbestpracticeinthisfieldareconstantlychanging.Asnewresearchandexperiencebroadenour
understanding,changesinresearchmethods,professionalpractices,ormedicaltreatmentmaybecomenecessary.
Practitionersandresearchersmustalwaysrelyontheirownexperienceandknowledgeinevaluatingandusingany
information,methods,compounds,orexperimentsdescribedherein.Inusingsuchinformationormethodstheyshouldbe
mindfuloftheirownsafetyandthesafetyofothers,includingpartiesforwhomtheyhaveaprofessionalresponsibility.
Tothefullestextentofthelaw,neitherthePublishernortheauthors,contributors,oreditors,assumeanyliabilityforany
injuryand/ordamagetopersonsorpropertyasamatterofproductsliability,negligenceorotherwise,orfromanyuseor
operationofanymethods,products,instructions,orideascontainedinthematerialherein.
LibraryofCongressCataloging-in-PublicationData
AcatalogrecordforthisbookisavailablefromtheLibraryofCongress
BritishLibraryCataloguing-in-PublicationData
AcataloguerecordforthisbookisavailablefromtheBritishLibrary
ISBN:978-0-12-804560-2
ForinformationonallButterworth-Heinemannpublications
visitourwebsiteathttps://www.elsevier.com/books-and-journals
Publisher:MaraConner
AcquisitionEditor:SonniniR.Yura
EditorialProjectManager:AnaClaudiaA.Garcia
ProductionProjectManager:BharatwajVaratharajan
Designer:MarkRogers
TypesetbyVTeX
Dedication
TomylovingwifeMarianaandmygrandchildrenNika,RukaandDominic.
DNN
Preface
Arobotmaynotinjureahuman
beingor,throughinaction,allow
ahumanbeingtocometoharm.
–IsaacAsimov
Humanoidrobotsarethemostuniversalmachinesdesignedsofar.Theirhuman-likeap-
pearance presumes that one day humanoid robots will become the ubiquitous helpers of
mankind as an embodiment of artificial intelligence (AI). Although still in its initial phase,
researchinthefieldofhumanoidroboticsisadvancingatafastpace,coveringtherebyadi-
versesetofproblems.Theresearchisbenefitingfrombutisalsocontributingtotechnological
achievements in such areas as self-driving transportation means in a complex environment
(e.g.inrelationtosensing,perception,andmotionplanning),naturallanguagecommunica-
tion(e.g.inrelationtopersonalassistance),andAIingeneral.Thisisalsotruefortheareas
inthefieldofthemechanicalsciencesandcontrol.
Characterized as universal, a humanoid robot is also an inherently complex machine. Its
controlarchitecturecomprisesahierarchicalstructure.Atthemiddlelevelsofthehierarchy,
kinematic,kinetostatic,anddynamicmodelshavetobeemployedtoensureappropriatecon-
trolofthemotionandtheflowofforces.Thisisachallengingproblemsincethemodelshave
to account for the relatively large number of actuated joints and the best way to simultane-
ouslycontrolthemwhileperformingavaryingnumberoftasks,withakinematicstructure
that varies frequently by forming closed kinematic chains and unforming them while the
robotestablishesnewcontactsandbreaksoldones.Themodelsalsohavetoaccountforthe
“floatingbase”andits“underactuation,”aswellasforthevaryingenvironmentconditions.
The aim of this book is the in-depth covering of a core set of problems related to the
modeling and the model-based control of humanoid robots. A large body of research ex-
ists in the robotics field that can support this goal. Kinematically redundant manipulators
and the related problem of multi-degree-of-freedom (DoF) resource allocation to manage
a number of robot tasks have been studied from the mid-1970s. A number of studies exist
onrobotswithclosedkinematicchains,referredtoasparallel-linkmanipulators.Studieson
structure-varyingroboticmechanisms,suchasmultifingeredhands,multileggedrobots,and
dual-/multiarmmanipulators,arealsoabundant.Underactuated,articulatedmultibodysys-
temsonafloatingbase,suchasfree-floatingspacerobots,manipulatorsonflexiblebases,and
macro-minimanipulators(i.e.asmallmanipulatorattachedtotheendofalargerarm)have
beenstudiedfromthemid-1980s.Thereareanumberofstudiesinthecloselyrelatedfieldof
constrainedmultibodysystems.Contactmodelingisalsoawell-establishedareaofresearch.
Theultimategoalistodesignahumanoidrobotcontrollerthatcanguaranteetheperfor-
mance of a broad variety of motion and force control tasks. This can only be done when a
xiii
xiv
PREFACE
whole-bodymodeloftherobotisemployed.Withawhole-bodymodel,thestructureofthe
controller becomes quite complex, though, as already noted. The control inputs then have
tobederivedwiththehelpofanoptimizationapproach.Anumberofhumanoidrobotcon-
trollersthatappearedrecentlyintheliteraturehavebeenbasedonthisapproach,whiletaking
advantageoftheexistinggeneraloptimizationsoftwarepackages.Becauseofthecomplexity,
however,real-timecontrolmaynotbefeasiblewiththisapproach.Ontheotherhand,witha
simplifiedmodelitispossibletoderivetheoptimalcontrolsolutionanalytically.Analytical
solutionsyieldtheadvantageofshortercomputationalcycles.Thesimplestpossiblemodelis
the(linear)invertedpendulummodel.Itwasproposedsometimeagotodealwiththetaskof
balancestabilizationonaflatground.Lateritwasshownthatthemodelcanbeenhancedby
adding a reaction wheel assembly component. A centroidal moment can then be generated
thatplaysanimportantrolewhendealingwithunknowndisturbancesappliedtothebodyof
therobotorthatappearwhiletherobotstepsoverirregularterrain,forexample.Analytical
optimal solutions for a whole-body model are quite rare, though. At present, the analytical
approachisindispensableformotiongenerationandcontrolinreal-time.
It is quite exciting that the research in the field of humanoid robotics can contribute to
otherareasaswell.Suchareasincludethebiomechanicsandmotorcontrolofhumanmove-
ment,physicaltherapy,thesportsciences,andphysics-basedanimationofarticulatedfigures.
Researchersinthesefieldscouldbenefitfromtheresultsdescribedinthisbookaswell.
Theauthorsofthisworkhavebeeninvolvedfordecadesintheareasofresearchthatcon-
stitute the foundation of humanoid robotics. It was a challenging task to draw on the past
results,toorganizeandreinterprettheminanattempttoexemplifytheirrolewhenapplied
tohumanoidrobots.Thisworkalsoincludesmanyimportantandup-to-dateresultsreported
by other researchers. From this point of view, the style resembles somewhat that of a refer-
ence.What makesitdifferent from ahandbook,though,is the revealingof someimportant
interrelationsbetweentheresultsinthedifferentareasofresearchandtheircontributionto
themaingoalofthiswork.
Tokyo,July2018 Theauthors
Acknowledgments
The following students from the Robotics Life Support laboratory at Tokyo City University
havecontributedtothiswork.
Shoichi Taguchi, Hiroki Sone, Ryosuke Mitsuhira, Masahiro Hosokawa, Sho Miyahara,
TakashiShibuya,andRyotaroHinataderivedsomeoftheequations.
TakuyaMimura,RyosukeMitsuhira,HironoriYoshino,KengoTamoto,MasahiroHosoka-
wa, Satoshi Shirai, Sho Miyahara, Takashi Shibuya, and Ryotaro Hinata helped to improve
thereadabilityofthetext.Theyalsoprovideddatafromsimulationsandexperiments,along
with Yuki Yoshida, Shogo Onuma, Kajun Nishizawa, Takuma Nakamura, Ryo Iizuka, and
RyotaYui.
Yuki Yoshida, Shogo Onuma, Shoichi Taguchi, Jie Chen, Keita Sagara, Tomonori Tani,
Ryosuke Mitsuhira, Takahide Hamano, Masamitsu Umekage, Kajun Nishizawa, Kei Koba-
yashi,TakashiShibuya,RyotaYui,YukiyaEndo,ShingoSakaguchi,TakayukiInoue,andYuki
Hidakacontributedwithfigurepreparation.
I appreciate the enormous patience of my wife to go carefully through the text and to
considerablyimprovetheEnglishtext.
DNN
xv
C H A P T E R
1
Introduction
1.1 HISTORICAL DEVELOPMENT
Forcenturies,imaginationhasbeenthedrivingforcebehindcreativity.Thisistrueespe-
ciallyforhumanoidrobotswiththeiroriginsbeingtraceablebacktoancientGreekmythol-
ogy.Thedrivetocreatehuman-likeartificialmachineshasneverceasedduringthecenturies
sincethen[32].Inmoderntimes,thetechnologicaladvancementsachievedbyengineershave
empowered them to bring into life the creations of the finest dreamers and science-fiction
writers.
The development of humanoid robots was pioneered in Japan with WABOT-1, designed
in1973bythelateprofessorIchiroKatoandhisstudentsatWasedauniversity[97].Foryears,
Japaneseengineersremainedthesolepursuersofthehumanoiddream.Theireffortseventu-
allypaidoff;thefascinatinghumanoidplatformdevelopedthroughouttheyearsbyHonda
Motor Co. and culminated in the world-renowned ASIMO humanoid [94]. In 1996, the im-
pactofunveilingASIMO’spredecessorP2tothepublicwasenormous.Thiseventuallyled
totheacceleratingdevelopmentofhumanoidrobottechnologywearewitnessingnowadays.
Governments in the developed countries are readily financing robot technology. This will
inevitably lead to an ever-accelerating cycle in synchrony with the existing exponential ad-
vancementsincomputingmachinery,asrevealedbyMoore’slaw.
Humanoid robots are complex machines. It took Honda’s engineers a decade to develop
a number of prototypes and to arrive at the P2 in 1996 [49,48]. It took another 15 years of
developmenttoarriveatthe“all-new”ASIMOin2011[99].Thisrobotisconsideredtobethe
world’smostadvancedhumanoidrobot,capableofrunningfast,veryhuman-like.Therobot
can also climb stairs, run backwards, hop on one or two legs continuously, and even walk
overuneventerrain.Thesephysicalcapabilitiesoftherobothavebeenachievedwithperfec-
tionist mechanical design, advanced sensors and actuator technology, and dynamic motion
control.BesidesimprovingASIMO’sphysicalcapabilities,Honda’sengineersalsodeveloped
therobot’sartificialintelligenceabilitiessuchasdecisionmakingformotionreplanningbased
onsensordatafusion,naturallanguage,gesture-basedcommunication,andothers[118].
Following Honda’s pioneering efforts, since the endof the 20th century a number of hu-
manoid robots have been built [33]. Humanoid robots have demonstrated such abilities as
drivingalifttruck[43]orabackhoe[44],pushing[39],lifting[93,83]ormovingbypivoting
[137] various heavy objects, opening and closing doors [110], pulling drawers [59], nailing
[127],liftingandcarryingobjectsincooperationwithhumans[27,16,2],andcooking[34].
HumanoidRobots 1
https://doi.org/10.1016/B978-0-12-804560-2.00008-0 Copyright©2019ElsevierInc.Allrightsreserved.
2
1. INTRODUCTION
Therobotsdevelopedsofar,however,areonlyprototypes.Theylacksufficientrobustness
tofunctioninthereal-lifeenvironment.TheDARPARoboticsChallenge(DRC)competition
[100]wasconceivedtoaddresssomeaspectsofthisproblem,inparticularwithregardtoan
extreme environment such as a disaster zone. Indeed, walking through uneven terrain and
rubbleanddrivingacarandgettingoutofitprovedtobetoodifficulttasksformostofthe
eighteen biped robots that took part in the competition. Seven of the biped teams used the
Atlasrobot[95],butonlyoneofthem,therunner-upRunningManofteamIHMC[96],could
complete all of the tasks. With nearly identical hardware (the Atlas robot teams used their
own lower-arm/hand designs), the robustness of balance maintenance can be identified as
being mainly a control issue. The competition outcome has clearly demonstrated, though,
thatenvironment-specificdesigncanalsoplayanimportantrole.Thewinnerandthethird-
place team (DRC-Hubo of team KAIST [61] and CHIMP of CMU team Tartan Rescue [98])
incorporated design elements that deviated from human-like forms, such as a biped-plus-
wheelandcaterpillar-basedlocomotion,respectively.
Biped humanoid robots cannot be expected to be commercialized in the near future. In
fact, Honda has announced the retirement of ASIMO [103]. The company is developing a
newhumanoidrobotwithadesigntosolvespecifictasksinadisaster-relatedenvironment
[138].Thecompanyalsorevealeditsplantoapplytheknow-how,accumulatedthroughout
the yearsof research,in suchareas as physicaltherapy and self-driving vehicles. Without a
doubt,pursuingmankind’sdreamwithcontinuingresearcheffortsinthefieldofhumanoid
roboticswillinevitablypayoffsomeday.
1.2 TRENDS IN HUMANOID ROBOT DESIGN
1.2.1 Human Likeness of a Humanoid Robot
Thedesignofauniversalhumanoidrobotcapableofperformingavarietyoftaskswithin
differentenvironmentsremainsanopenissue.Thecommon“formfollowsfunction”design
principlecanleadtobothadvantagesanddisadvantagesinhumanoidrobotdesign,asdis-
cussed by Stanford university professor Bernard Roth [117]. The external appearance of a
humanoid robot, including the way it moves, plays an important role for its acceptance in
society,asnotedbyprofessorMasahiroMoriofTokyoInstituteofTechnology[79].Attempts
ofquantifyingthehuman-likenessofahumanoidrobotarepresentedin[104,146].
As an example, consider one of the main functions of a biped robot: walking on a level
ground. Walking has been realized with the help of a 3D linear inverted pendulum (LIP)
model[62].This,however,hasresultedin“crouched”gaitswithoutstraighteningtheknees
as in the typical erect gait of the modern-times human. Besides the external appearance
problem of the “crouched” robot gait, there is also a functional problem related to gait ef-
ficiency. As noted in a study on the bipedal walking of the early hominid Australopithecus
afarensis[20],netenergyabsorptionispredictedforthe“bent”joints,whichwouldhavere-
sulted in increased heat load. Indeed, erect (straight-leg) walking is a characteristic of the
mostenergy-efficientmachinegait:thatofapowerlessbipeddescendingaslope,knownas
passive dynamic walk (PDW) [74]. The bent-knee gait in powered bipeds has been identi-
fiedasaproblemin[119,92]andtackledlateronin[70,80,64,124,63,38,71,41,139].Improved