Tothisend, thenational-localJointHumanoidRobotInnovationCentershowcaseditsprogressbuildingtherobot's capabilitiesanddesignconceptsin a livedemonstration.
Next, intermsofdevelopment, Qinglongis a creationofHumanoidRobotsShanghaiLimited, an R&D institutionbackedbyleadingindustryenterpriseswithapproximately 140 millionUSDinfunding.
Thisinitialphasewillallowthecompanytofine-tunetherobot's capabilitiesin a controlledenvironment, andonceproveneffective, Teslawillthenbegincommercialsalesoftherobot, makingitavailableto a broadermarketsometimein 2025.
Thisextensivedatasetcomprisesoveronemillionframescapturedfrommultipleperspectives, offering a richresourceforresearchersdelvingintothecomplexitiesofhowhumansmanipulateobjectswiththeirhands.
AndtheHOT3D datasetis a treasuretroveofover 800 minutesofegocentricvideorecordings, withtheserecordingsbeingsynchronizedacrossmultipleperspectivesandincludinghigh-quality 3D poseannotationsforbothhandsandobjects.
Thedatasetalsofeatures 3D objectmodels, withphysically-basedrenderingmaterials, 2D boundingboxes, gazesignals, and 3D scene-pointcloudsgeneratedfromsimultaneouslocalizationandmappingtechniques.
Inthesecondstep, TextureGencaneitherrefinethetextureoftheobjectgeneratedinthefirststepbasedontheoriginaldescription, orcreate a textureforanyunstructured 3D meshusingnewspecifications.
Thesemodelsareoptimizedwithsynthetic 3D datatoenablemulti-viewgenerationinbothimageandUVspace, whichsignificantlyenhancesthetexturequalityofthefinalproduct.
Metaclaimsthat 3DGenisthreetosixtytimesfasterthancomparablesystems, a significantleapthatcoulddrasticallyreducethetimeandeffortrequiredfor 3D contentcreation.