全能型机器人的PPT.pdfVIP

LearningFine-GrainedBimanualManipulationwith

Low-CostHardware

TonyZ.Zhao1VikashKumar3SergeyLevine2ChelseaFinn1

1StanfordUniversity2UCBerkeley3Meta

ExampleTeleoperatedSkills

threadzipcabletieNISTboard#2jugglepingpong

ExampleLearnedPolicy

openlidslotbatterymanipulatetape

Fig.1:ALOHA:ALow-costOpen-sourceHardwareSystemforBimanualTeleoperation.Thewholesystemcosts$20kwithoff-the-shelf

robotsand3Dprintedcomponents.Left:Theuserteleoperatesbybackdrivingtheleaderrobots,withthefollowerrobotsmirroringthemotion.

Right:ALOHAiscapableofprecise,contact-rich,anddynamictasks.Weshowexamplesofbothteleoperatedandlearnedskills.

Abstract—Finemanipulationtasks,suchasthreadingcabletiescupfrombelowandpriesthelidopen.Eachofthesesteps

orslottingabattery,arenotoriouslydifficultforrobotsbecauserequireshighprecision,delicatehand-eyecoordination,and

theyrequireprecision,carefulcoordinationofcontactforces,richcontact.Millimetersoferrorwouldleadtotaskfailure.

andclosed-loopvisualfeedback.Performingthesetaskstypically

requireshigh-endrobots,accuratesensors,orcarefulcalibration,Existingsystemsforfinemanipulationuseexpensiverobots

whichcanbeexpensiveanddifficulttosetup.Canlea