Qaababka Ficil-aragga-Luuqada ee Kediska Robotics

Jawaabaha iyo sharraxaadaha waxay soo baxayaan markaad qaadanayso kediska.

1. Waa maxay saddexda nooc ee wax-soo-saarka/wax-soo-saarka ayaa qeexaya qaabka-Hadalka-Luuqadda-Hawlgalka (VLA)?
   • Sawirada kamarada, barashada luqadda, iyo ficilada robotka
   • Maqal, qoraal, iyo sawirro
   • Logyada dareemayaasha, GPS, iyo xaaladda batteriga
   • Koodhka, soo-saarka isku-darka, iyo natiijooyinka tijaabada
2. Sidee Google DeepMind's RT-2 u matashay ficillada robot-ka si koronto-bixiye u soo saaro?
   • Sida xididada sabbaynaya ee joogtada ah oo kaliya
   • Sida calaamado ficil ah oo qarsoodi ah, oo la mid ah calaamadaha qoraalka
   • Sida fariinta ROS XML
   • Sida qiyamka abaalmarinta ee xoojinta barashada
3. Waa maxay macnaha 'wareejinta degdega ah' marka loo eego macnaha guud ee VLA-yada?
   • Robotku aad buu u kululaadaa wuuna xiraa
   • Laba robots ayaa isweydaarsada barnaamijyadooda
   • Aqoonta cabbirka mareegaha ee tababarka hore waxay ka caawisaa robot-ku inuu qabsado walxaha iyo tilmaamaha cusub
   • Ficilada waxaa la xisaabiyaa iyada oo aan la helin shabakad neerfaha ah
4. Waa maxay faa'iidada muhiimka ah ee pi-0's faafinta/qulqulka-ku-habboonaanta madaxa ficilka marka la barbar dhigo halbeegga ficilka gaarka ah?
   • Waxay meesha ka saaraysaa baahida kamaradaha
   • Kaliya waxay ku shaqeysaa jilitaanka
   • Waxay meesha ka saaraysaa barashada luqadda
   • Waxay soo saartaa jajabyo siman oo soo noqnoqonaya oo sarreeya si loo xakameeyo qalafsanaan badan
5. Waa maxay sababta xog-ururinta Furan ee X-Embodiment ay muhiim ugu tahay VLA-yada?
   • Waxay ku ururisaa xogta robot-ka noocyo badan oo robot ah si ay awood ugu siiso guud ahaan qaab-dhismeedka
   • Waa halbeegga kaliya ee qoraalka sawirka
   • Waxa ay kaydisaa xogta hal robot oo kaliya
   • Waa corpus qoraal-luqad kaliya ah
6. Which data sources did DeepMind combine when co-fine-tuning RT-2?
   • Only a fixed list of hand-written joint commands
   • Robot demonstration data and web data with a pretrained vision-language model
   • Only audio recordings without robot observations
   • Only random action tokens with no examples
7. What kinds of commands can an RT-2-style action representation encode?
   • Only natural-language explanations with no robot controls
   • A fixed video caption for each camera frame
   • Discretized robot motion and gripper commands represented as tokens
   • A robot’s hardware serial number
8. How broad was the RT-2 evaluation of emergent capabilities reported by Google DeepMind?
   • Exactly 12 trials
   • One trial for every robot type worldwide
   • No physical robot trials
   • More than 6,000 robotic trials