Maswali yanayounganishwa na mwongozo · Kati Kiwango

Sera ya Usambazaji kwa Maswali ya Kudhibiti Roboti

Tazama jinsi Sera ya Usambazaji inavyoweka deno mfuatano wa vitendo kwa udhibiti wa visuomotor na kupanga upya kutoka kwa uchunguzi mpya wakati wa utekelezaji.

Njia za mwongozo zinazohusianaSera ya Usambazaji kwa Udhibiti wa Roboti
Swali 1 ya 8

Sera ya Usambazaji hutoa nini katika kila hatua ya uamuzi?

Hakiki kila swali8 maswali

Majibu na maelezo huonekana unapojibu maswali.

  1. Sera ya Usambazaji hutoa nini katika kila hatua ya uamuzi?
    • Mlolongo mfupi (chunk) wa vitendo vya siku zijazo
    • Amri moja ya papo hapo ya gari
    • Ufafanuzi wa lugha ya asili
    • Ishara ya zawadi kwa mafunzo
  2. Sera ya Usambazaji hukopa kutoka kwa AI ya picha?
    • Uchoraji wa pikseli kwa mikono
    • Kutoa sauti mara kwa mara kwa kelele nasibu
    • Algorithms ya maumbile
    • Miti ya maamuzi
  3. How can a generative action policy represent a task with several valid solutions?
    • It must choose the same fixed action for every observation
    • It removes the need for demonstrations or action data
    • It can model multiple plausible action sequences instead of regressing to their average
    • It guarantees higher success than every older controller
  4. Wakati wa mafunzo, mtandao katika Sera ya Usambazaji unafunzwa kutabiri nini?
    • Neno linalofuata katika sentensi
    • Kiwango cha betri ya roboti
    • Mpangilio wa kukaribia aliye na kamera
    • Kelele ambayo iliongezwa kwa mfuatano wa kitendo ulioonyeshwa
  5. 'Receding-horizon' ni nini kupanga upya katika Sera ya Usambazaji?
    • Kutabiri sehemu ya hatua lakini kupanga upya kwa uchunguzi mpya kabla ya kukimaliza
    • Kuendesha roboti mara moja tu kisha kuacha
    • Inapuuza ingizo zote za kihisi
    • Mafunzo bila maandamano
  6. Which architecture did the Diffusion Policy paper introduce for modeling temporal action sequences?
    • A static image classifier with no temporal model
    • A time-series diffusion transformer
    • A text-only recurrent language model
    • A fixed lookup table of robot motions
  7. What observations can condition a visuomotor Diffusion Policy?
    • Only a text prompt, with no robot or scene data
    • The names of the training researchers
    • Visual observations and the robot’s state
    • A future action sequence labeled as correct at inference time
  8. How broad was the Diffusion Policy paper’s benchmark evaluation?
    • One task on a single robot benchmark
    • Hundreds of tasks across every robot platform
    • Only simulations with no manipulation tasks
    • Twelve tasks across four robot-manipulation benchmarks