Amarobhothi ayisebenzisa kanjani i-AI? [Ividiyo Nombuzo]

Impendulo emfushane: Amarobhothi asebenzisa i-AI ukuqhuba umjikelezo oqhubekayo wokuzwa, ukuqonda, ukuhlela, ukwenza, nokufunda, ukuze akwazi ukuhamba nokusebenza ngokuphephile ezindaweni ezigcwele, ezishintshayo. Lapho izinzwa zithola umsindo noma ukuzethemba kwehla, izinhlelo eziklanywe kahle ziyehla, ziyeke ngokuphepha, noma zicele usizo kunokuqagela.

Izinto ezibalulekile okufanele uzicabangele:

I-Autonomy loop: Yakha izinhlelo ezizungeze umqondo–ukuqonda–uhlelo–ukwenza–ukufunda, hhayi imodeli eyodwa.

Ukuqina: Umklamo wokukhanya, ukuxinana, ukushelela, kanye nabantu abahamba ngokungalindelekile.

Ukungaqiniseki: Khipha ukuzethemba bese ukukusebenzisa ukuqala ukuziphatha okuphephile nokunobulungiswa.

Amalogi okuphepha: Qopha izenzo kanye nomongo ukuze ukwehluleka kuhlolwe futhi kulungiswe.

I-Hybrid stack: Hlanganisa i-ML nemingcele yefiziksi kanye nokulawula kwakudala ukuze kube nokuthembeka.

Ngezansi umbono ophelele wokuthi i-AI ivela kanjani ngaphakathi kwamarobhothi ukuze asebenze kahle.

Izihloko ongase uthande ukuzifunda ngemva kwalesi:

🔗 Lapho amarobhothi ka-Elon Musk esongela imisebenzi
Lokho amarobhothi kaTesla angakwenza nokuthi yiziphi izindima ezingashintsha.

🔗 Iyini irobhothi le-humanoid AI
Funda ukuthi amarobhothi angumuntu aqonda kanjani, anyakaze, futhi alandele kanjani imiyalelo.

🔗 Yimiphi imisebenzi ezothathelwa indawo yi-AI
Izindima ezivezwa kakhulu ku-automation kanye namakhono ahlala ewusizo.

🔗 Imisebenzi yobuhlakani bokwenziwa kanye nemisebenzi yesikhathi esizayo
Izindlela zomsebenzi ze-AI zanamuhla kanye nendlela i-AI eguqula ngayo izitayela zomsebenzi.

Amarobhothi ayisebenzisa kanjani i-AI? Imodeli yengqondo esheshayo

Amarobhothi amaningi asebenzisa i-AI alandela iluphu efana nale:

• Inzwa 👀: Amakhamera, imakrofoni, i-LiDAR, izinzwa zamandla, ama-encoder amasondo, njll.

• Qonda 🧠: Thola izinto, ulinganisele indawo, uqaphele izimo, ubikezele ukunyakaza.

• Hlela 🗺️: Khetha imigomo, bala izindlela eziphephile, hlela imisebenzi.

• Yenza 🦾: Khiqiza imiyalo yemoto, bamba, gingqa, balance, gwema izithiyo.

• Funda 🔁: Thuthukisa ukuqonda noma ukuziphatha okuvela kudatha (ngezinye izikhathi ku-inthanethi, ngokuvamile ungaxhunyiwe ku-inthanethi).

Iningi le-"AI" eliyirobhothi empeleni liyinqwaba yezingcezu ezisebenza ndawonye -ukuqonda, ukulinganisa isimo, ukuhlela, kanye nokulawula- okuhlanganisa ukuzimela.

Iqiniso elilodwa "lensimu" elisebenzayo: ingxenye enzima ngokuvamile akukhona ukwenza irobhothi lenze into kanye ku-demo ehlanzekile - kuwukwenza lenze into efanayo elula ngokwethembeka lapho izibani zishintsha, amasondo eshelela, phansi kucwebezela, amashalofu esukile, futhi abantu behamba njenge-NPC ezingalindelekile.

Yini eyenza ubuchopho obuhle be-AI kurobhothi

Ukusethwa kwe-AI yerobhothi eqinile akufanele kube okuhlakaniphile nje kuphela - kufanele kube okuthembekile ezindaweni ezingalindelekile, zangempela.

Izici ezibalulekile zifaka:

• Ukusebenza kwesikhathi sangempela ⏱️ (ukufaneleka kwesikhathi kubalulekile ekwenzeni izinqumo)

• Ukuqina kwedatha engcolile (ukukhanya, umsindo, imfuhlumfuhlu, ukufiphala kokunyakaza)

• Izindlela zokwehluleka ezinhle 🧯 (yehlisa ijubane, yeka ngokuphepha, cela usizo)

• Izinto ezinhle zokuqala + ukufunda okuhle (ifiziksi + imikhawulo + i-ML - hhayi nje "ama-vibes")

• Ikhwalithi yokuqonda elinganiswayo 📏 (ukwazi ukuthi izinzwa/amamodeli awonakele nini)

Amarobhothi amahle kakhulu avame ukuba yilawo angenza iqhinga elimangalisayo kanye, kodwa yilawo angenza imisebenzi eyisicefe usuku nosuku.

Ithebula Lokuqhathanisa Lamabhulokhi Okwakha E-AI Ejwayelekile Yerobhothi

Umbono: Indlela Amarobhothi Aguqula Ngayo Idatha Yezinzwa Ezingavuthiwe Ibe Yincazelo

Ukuqonda yilapho amarobhothi ephendula khona imifudlana yezinzwa ibe yinto angayisebenzisa ngempela:

• Amakhamera → ukuqashelwa kwezinto, ukulinganisa ukuma, ukuqonda indawo

• I-LiDAR → ibanga + i-geometry yezithiyo

• Amakhamera okujula → isakhiwo se-3D kanye nendawo ekhululekile

• Ama-Microphone → Izimpawu zenkulumo nomsindo

• Izinzwa zamandla/i-torque → ukubamba okuphephile kanye nokubambisana

• Izinzwa zokuthinta → ukutholwa kokushelela, imicimbi yokuxhumana

Amarobhothi athembele ku-AI ukuphendula imibuzo efana nale:

• "Yiziphi izinto eziphambi kwami?"

• "Ingabe lowo ngumuntu noma umodeli?"

• "Iphi isibambo?"

• "Ingabe kukhona okungihambelayo?"

Imininingwane ecashile kodwa ebalulekile: izinhlelo zokuqonda kufanele ziveze ukungaqiniseki (noma i-proxy yokuzethemba), hhayi nje impendulo ethi yebo/cha-ngoba ukuhlela okulandelayo kanye nezinqumo zokuphepha kuncike ekutheni irobhothi liqiniseke kangakanani

Ukuhlela Indawo Nokumamephu: Ukwazi Ukuthi Ukuphi Ngaphandle Kokwethuka

Irobhothi lidinga ukwazi ukuthi lizosebenza kuphi kahle. Lokhu kuvame ukuphathwa nge -SLAM (Indawo kanye Nokumapha Ngesikhathi Esifanayo): ukwakha imephu ngenkathi kulinganiswa ukuma kwerobhothi ngesikhathi esifanayo. Ezakhiweni zakudala, i-SLAM iphathwa njengenkinga yokulinganisa okungenzeka, ngemindeni evamile ehlanganisa izindlela ezisekelwe ku-EKF kanye nezisekelwe ku-particle-filter. [1]

Irobhothi ngokuvamile lihlanganisa:

• I-odometry yamasondo (ukulandelela okuyisisekelo)

• Ukufanisa ukuskena kwe-LiDAR noma izimpawu ezibonakalayo

• Ama-IMU (ukujikeleza/ukusheshisa)

• I-GPS (ngaphandle, enemikhawulo)

Amarobhothi awakwazi ukuhlala endaweni efanele ngokuphelele - ngakho ama-stack amahle asebenza njengabantu abadala: alandelela ukungaqiniseki, athole ukuzulazula, bese ebuyela ekuziphatheni okuphephile lapho ukuzethemba kwehla.

Ukuhlela Nokwenza Izinqumo: Ukukhetha Okufanele Ukwenze Okulandelayo

Uma irobhothi selinomfanekiso osebenzisekayo womhlaba, kudingeka linqume ukuthi lizokwenzani. Ukuhlela kuvame ukubonakala ngezingqimba ezimbili:

• Ukuhlela kwendawo (ukusabela okusheshayo) ⚡
  Gwema izithiyo, yehlisa ijubane eduze kwabantu, landela imizila/imihubhe.

• Ukuhlela komhlaba wonke (isithombe esikhulu) 🧭
  Khetha izindawo oya kuzo, hamba uzungeze izindawo ezivinjiwe, hlela imisebenzi.

Empeleni, yilapho irobhothi liphendula khona elithi “Ngicabanga ukuthi ngibona indlela ecacile” libe yimiyalo yokunyakaza engeke ifinyelele ekhoneni leshalofu—noma ingene endaweni yomuntu siqu.

Ukulawula: Ukuguqula Amacebo Abe Ukunyakaza Okushelelayo

Izinhlelo zokulawula ziguqula izenzo ezihleliwe zibe yisenzo sangempela, kuyilapho zibhekana nezinkathazo zomhlaba wangempela ezifana nalezi:

• Ukungqubuzana

• Izinguquko zomthwalo

• Amandla adonsela phansi

• Ukulibaziseka kwezimoto kanye nokusabela okubi

Amathuluzi avamile afaka phakathi i-PID, ukulawula okusekelwe kumodeli, ukulawula kokubikezela kwemodeli, kanye ne-inverse kinematics yezingalo - okungukuthi, izibalo eziphendula "ukubeka i-gripper lapho" ekunyakazeni kwamalunga. [2]

Indlela ewusizo yokucabanga ngakho:
Ukuhlela kukhetha indlela.
Ukulawula kwenza irobhothi liyilandele ngaphandle kokuzamazama, ukudubula ngokweqile, noma ukudlidliza njengenqola yokuthenga ene-caffeine.

Ukufunda: Indlela Amarobhothi Athuthuka Ngayo Esikhundleni Sokuhlelwa Kabusha Kuze Kube Phakade

Amarobhothi angathuthuka ngokufunda idatha kunokuba alungiswe ngesandla ngemva kokushintsha kwendawo.

Izindlela zokufunda ezibalulekile zifaka:

• Ukufunda okuqondisiwe 📚: Funda ezibonelweni ezinelebula (isib., “lena yiphalethi”).

• Ukufunda okuqondiswa nguwe 🔍: Funda isakhiwo kusuka kudatha eluhlaza (isb., ukubikezela ozimele besikhathi esizayo).

• Ukufunda kokuqinisa 🎯: Funda izenzo ngokukhulisa izimpawu zomvuzo ngokuhamba kwesikhathi (ngokuvamile ezihlanganiswe nama-ejenti, izindawo, kanye nokubuyiselwa kwemali). [3]

Lapho i-RL ikhanya khona: ukufunda ukuziphatha okuyinkimbinkimbi lapho ukuklama ngesandla isilawuli kuba buhlungu khona.
Lapho i-RL iba mnandi khona: ukusebenza kahle kwedatha, ukuphepha ngesikhathi sokuhlola, kanye nezikhala ze-sim-to-real.

Ukusebenzisana Kwabantu Namarobhothi: I-AI Esiza Amarobhothi Asebenze Nabantu

Kumarobhothi asemakhaya noma ezindaweni zokusebenza, ukuxhumana kubalulekile. I-AI ivumela:

• Ukuqashelwa kwenkulumo (umsindo → amagama)

• Ukutholwa kwenhloso (amagama → incazelo)

• Ukuqonda ngezandla (ukukhomba, ulimi lomzimba)

Lokhu kuzwakala kulula uze ukuthumele: abantu abavumelani, izindlela zokukhuluma ziyahlukahluka, amakamelo anomsindo, futhi "lapho" akuyona ifreyimu ehlanganisiwe.

Ukwethemba, Ukuphepha, kanye nokuthi “Ungabi Othusayo”: Ingxenye Engajabulisi Kakhulu Kodwa Ebalulekile

Amarobhothi ayizinhlelo ze-AI ezinemiphumela ebonakalayo, ngakho-ke ukwethembana kanye nemikhuba yokuphepha akufanele kube yinto ecatshangelwe kamuva.

I-scaffolding yokuphepha ewusizo ivame ukufaka:

• Ukuqapha ukuzethemba/ukungaqiniseki

• Ukuziphatha okunobulungiswa lapho ukuqonda kwehla

• Izenzo zokubhalisa zokulungisa amaphutha nokuhlolwa

• Sula imingcele yalokho irobhothi elingakwenza

Indlela ewusizo ephezulu yokuhlela lokhu ukuphathwa kwezingozi: ukubusa, ukumaka izingozi, ukuzilinganisa, nokuziphatha kuyo yonke impilo - okuhambisana nendlela i-NIST ehlela ngayo ukuphathwa kwezingozi ze-AI kabanzi. [4]

Umkhuba “Wemodeli Enkulu”: Amarobhothi Asebenzisa Amamodeli Esisekelo

Amamodeli ayisisekelo aqhubekela phambili ekuziphatheni kwamarobhothi okuhloswe ngakho-ikakhulukazi lapho ulimi, umbono, kanye nesenzo kuhlanganiswa ndawonye.

Isibonelo esisodwa sesiqondiso -vision-language-action (VLA) , lapho uhlelo luqeqeshwa khona ukuxhuma lokho olukubonayo + lokho olutshelwa ukuba lukwenze + ukuthi yiziphi izinyathelo okufanele luzithathe. I-RT-2 iyisibonelo esicashunwe kabanzi salolu hlobo lwendlela. [5]

Ingxenye ethokozisayo: ukuqonda okuguquguqukayo, okusezingeni eliphezulu.
Ukuhlolwa kwamaqiniso: ukuthembeka komhlaba ongokoqobo kusadinga izindlela zokuqapha - ukulinganisa okuvamile, imikhawulo yokuphepha, kanye nokulawula okulondoloziwe akupheli ngoba nje irobhothi likwazi "ukukhuluma ngokuhlakanipha."

Amazwi Okugcina

Ngakho-ke, Amarobhothi ayisebenzisa kanjani i-AI? Amarobhothi asebenzisa i-AI ukubona, ukulinganisa isimo (ngikuphi?), ukuhlela, nokulawula- futhi ngezinye izikhathi afunde kudatha ukuze athuthuke. I-AI ivumela amarobhothi ukuthi abhekane nobunzima bezindawo eziguquguqukayo, kodwa impumelelo incike ezinhlelweni ezinokwethenjelwa, ezilinganiswayo ezinokuziphatha kokuphepha kuqala.

Isibonelo sangempela: Ukwakha umsizi we-AI werobhothi lokugcina impahla

Isimo

Cabanga ngendlu encane yokugcina izinto esebenzisa irobhothi elihambayo elizimele ukuhambisa amathothi avaliwe kusuka emabhentshini okupakisha kuya endaweni yokuthumela. Irobhothi akudingeki "liqonde konke". Lidinga ukwenza umsebenzi owodwa ngokwethembeka: ukuqoqa ithothi, ukuhamba endaweni ehlanganyelwe, ukugwema abantu namaloli ephalethi, bese lima ngokuphepha lapho ukuzethemba kwehla.

I-AI stack ingahlanganisa umbono wekhompyutha, i-LiDAR, i-SLAM, ukuhlela indlela, ukugwema izithiyo, kanye nemiyalelo eyisisekelo yolimi evela kubasebenzi. Umphathi angase athi, “Thatha leli thuluzi uye nalo endaweni yokuthumela i-bay 3,” kodwa irobhothi lisadinga imithetho eqinile yokuphepha ngaphansi kongqimba lolimi.

Lesi yisibonelo esinamandla ngoba sibonisa i-AI yerobhothi isebenza njengesitaki esisebenzayo, kunokuba kube yimodeli eyodwa enkulu eqagelayo.

Lokho okudingwa umsizi

Ukusetha kuzodinga:

• Imephu yendawo yokugcina impahla, kufaka phakathi amabhentshi okupakisha, izindawo zokuthumela impahla, izindawo ezingadingi ukuhamba, izindawo zokushaja, kanye nezindlela ezincane

• Idatha yekhamera noma yekhamera ejulile yokubona amathoyizi, abantu, izimpawu zephansi, kanye nemizila evinjiwe

• I-LiDAR noma enye inzwa yebanga yokuthola izithiyo

• Ama-encoder amasondo kanye nedatha ye-IMU yokwenza kube okwasendaweni

• Imithetho yemikhawulo yesivinini, ibanga lokuma, kanye nokuziphatha okuphephile kubantu

• Uhlu lwemisebenzi oluvela ohlelweni lwendawo yokugcina impahla, njenge-tote ID, indawo yokuthathwa, kanye nendawo oya kuyo

• Amalogi aqopha umzila, amaphuzu okuzethemba, ukuma, cishe ukuphuthelwa, kanye nokungenelela kwabantu

Iphuzu elibalulekile ukuthi imfundo yolimi akufanele ibe yiyo yodwa ungqimba lokulawula. Irobhothi lingayamukela imiyalo enobungani nabantu, kodwa ukunyakaza kufanele kusabuswe yimingcele yemephu, ukuhlela, ukulawula, kanye nokuphepha.

Isibonelo semiyalelo

Imiyalelo elula yokusebenza yomsizi werobhothi ingabukeka kanje:

Ungumsizi womsebenzi werobhothi elihambayo lendawo yokugcina impahla. Umsebenzi wakho ukuguqula izicelo zabasebenzi zibe imisebenzi yokuhamba ephephile. Dala umsebenzi kuphela uma indawo yokuthathwa, indawo oya kuyo, kanye ne-tote ID kucacile. Uma kukhona imininingwane engekho, buza umbuzo owodwa omfushane wokucacisa. Ungalokothi usungule izindawo. Uma umzila uvinjiwe, khetha umzila ohlukile ovunyiwe uma ukhona. Uma ukuzethemba kokubona kwehla ngaphansi komkhawulo ophephile, yehlisa ijubane, yima, noma ucele usizo lomuntu. Bhala phansi yonke i-pickup ehlulekile, umzila ovinjiwe, ukuma okuphuthumayo, kanye nokushintsha ngesandla.

Isicelo sabasebenzi singaba:

Susa i-tote T-1842 kusuka ebhentshini lokupakisha 6 uye endaweni yokuthumela 3.

Impendulo enhle ingaba:

Umsebenzi wamukelekile: qoqa i-tote T-1842 ebhentshini lokupakisha 6 bese uyiletha endaweni yokuthumela 3. Indlela ekhethiwe nge-aisle B. Ijubane linqunyelwe eduze kwendawo yokuwela abahamba ngezinyawo. Uma i-aisle B ivinjiwe, sebenzisa umzila ovunyiwe wokubuyela emuva C.

Impendulo embi ingaba:

Impela, ngizoyiyisa ekuthumeleni.

Inguqulo embi ayicacile kahle. Ayiqinisekisi i-tote ID, indawo yokuthathwa, indawo oya kuyo, umzila, noma ukuziphatha kokuphepha.

Indlela yokuyihlola

Ngaphambi kokuvumela irobhothi ukuthi lisebenze endaweni ebukhoma, lihlole ngohlu lokuhlola oluncane:

• Cela ukuthi ihambise ithoyizi elinemininingwane ephelele

• Cela ukuthi ihambise ithoyizi ngaphandle kokunikeza indawo yokuthumela

• Beka isithiyo esifana nomuntu endleleni

• Hambisa umaka weshelufu bese uhlola ukuthi ukuzethemba kwendawo kuyehla yini

• Dala ukukhanya phansi bese uhlola ukuthi ukuzethemba kokubona kuyashintsha yini

• Vimba indlela oyithandayo bese uhlola ukuthi ikhetha umzila wokubuyela emuva ovunyelwe yini

• Cela indawo engekho bese uhlola ukuthi iyenqaba yini esikhundleni sokuqagela

• Buyekeza ilogi ngemva kokugijima ngakunye ukuqinisekisa ukuthi ukuma, ukushintsha imizila, kanye nokweqa kuqoshiwe

Umgomo akukhona nje ukuthi “ingabe irobhothi lifikile?” Umbuzo ongcono uwukuthi: “Ingabe liziphathe ngokuphepha nangokubikezela lapho imvelo iba nokungaqiniseki?”

Umphumela

Umphumela obonisayo: ngokusekelwe esikhathini imisebenzi engu-20 yokuthutha amathoyizi endaweni encane yokuhlola izimpahla.

Ngaphambi kokusebenzisa umsebenzi werobhothi, umuntu ogijimayo wayethatha isilinganiso semizuzu emi-4 nemizuzwana engama-30 ngokunyakaza ngakunye, okuhlanganisa nokuhamba ebuyela ebhentshini lokupakisha. Ngemva kokwethula irobhothi ukuze lidlulisele kalula ithoyizi eliqondile, isikhathi sokuphatha umuntu sehle saba cishe yimizuzwana engama-50 ngomsebenzi ngamunye, ikakhulukazi ukulayisha ithoyizi nokuqinisekisa umsebenzi.

Lokho kuzokonga cishe imizuzu emi-3 nemizuzwana engama-40 ngokunyakaza ngakunye. Kuma-tote angu-80 ngosuku, isikhathi esilinganiselwe sokulondolozwa singaba cishe imizuzu engama-293, noma ngaphansi kwamahora abasebenzi angu-4.9 ngosuku.

Ukuhlolwa kokuphepha esivivinyweni esifanayo kufanele kulandelwe ngokwehlukana. Isibonelo:

• Imisebenzi engu-20 kwengu-20 ifinyelele endaweni efanele

• Imicimbi emithathu yemizila evinjiwe iphathwe ngokushintsha umzila okuvunyelwe

• Izehlakalo ezimbili zokungazethembi zibangele ukuma okuphephile

• Izindawo ezingavunyiwe eziyi-0 zamukelwe

• Kuqagelwe ukuthi ama-ID e-tote alahlekile ayi-0

Lezi zinombolo ziyisibonelo, hhayi isimangalo mayelana nanoma yimuphi umkhiqizo werobhothi othize. Ithimba lingaqinisekisa umphumela ngokubeka isikhathi semisebenzi ngaphambi nangemva kokusetshenziswa, ukubala ukukhishwa ngesandla, ukubuyekeza amalogi omzila, nokuhlola ukulethwa okuhlulekile.

Yini engase ihambe kabi

Ukwehluleka okuvame kakhulu ukunikeza irobhothi inkululeko eningi kakhulu. Imodeli yolimi ingase iqonde imiyalelo, kodwa lokho akusho ukuthi kufanele ithembeke ekusunguleni izindlela, inganaki amaphuzu okuzethemba, noma inqume ukuthi yini "cishe ephephile".

Ezinye izinkinga ezingokoqobo zifaka:

• Amamephu aphelelwe yisikhathi ngemva kokususwa kwamashalofu noma amabhentshi

• Ukukhanya okubi noma phansi okukhanyayo okudidayo amamodeli okubona

• Abasebenzi abasebenzisa amagama endawo angakahleleki irobhothi elingawazi

• Ama-tote ID angekho abangela ukuthi uhlelo lukhethe into engalungile

• Ukungena ngemvume okubuthakathaka, okwenza kube nzima ukuphenya ngokugeja

• Ukubiza ngaphezulu ukusebenza ngaphandle kokulinganisa ukugijima okuhlulekile kanye nokungenelela kwabantu

Umthetho onengqondo ulula: uma irobhothi lingaqiniseki, kufanele libe nokugcina izinto ngendlela ehlelekile, hhayi ukuqamba izinto ngendlela enobuchule.

Ukudla okuwusizo

Ukusethwa kwe-AI yerobhothi eqinile kwakhiwe ngomsebenzi omncane, okokufaka okucacile, ukuziphatha kokuphepha okulinganisekayo, kanye nokubuyela emuva okuthembekile. "Ukuhlakanipha" akukhona nje ukuqaphela izinto noma ukulandela imiyalelo. Kuwukwazi ukuthi kufanele uthuthe nini, ukuthi kufanele wehlise ijubane nini, ukuthi kufanele ume nini, nokuthi kufanele ucele usizo nini.

Imibuzo Evame Ukubuzwa

Amarobhothi ayisebenzisa kanjani i-AI ukuze asebenze ngokuzimela?

Amarobhothi asebenzisa i-AI ukuze aqhube i-autonomic loop eqhubekayo: ukuzwa umhlaba, ukuhumusha okwenzekayo, ukuhlela isinyathelo esilandelayo esiphephile, ukusebenza ngezimoto, nokufunda kudatha. Empeleni, lokhu kuyinqwaba yezingxenye ezisebenza ngokubambisana kunokuba kube yimodeli eyodwa "yomlingo". Inhloso ukuziphatha okuthembekile ezindaweni ezishintshayo, hhayi i-demo yesikhathi esisodwa ngaphansi kwezimo eziphelele.

Ingabe i-robot AI iyimodeli eyodwa nje noma i-full autonomic stack?

Ezinhlelweni eziningi, i-robot AI iyinqwaba egcwele: ukuqonda, ukulinganisa isimo, ukuhlela, kanye nokulawula. Ukufunda komshini kusiza ngemisebenzi efana nombono nokubikezela, kuyilapho imikhawulo yefiziksi kanye nokulawula kwakudala kugcina ukunyakaza kuzinzile futhi kubikezelwa. Ukufakwa okuningi kwangempela kusebenzisa indlela ehlanganisiwe ngoba ukuthembeka kubaluleke kakhulu kunokuhlakanipha. Yingakho ukufunda "okuzwakalayo kuphela" kuvame ukusinda ngaphandle kwezilungiselelo ezilawulwayo.

Yiziphi izinzwa namamodeli okuqonda amarobhothi e-AI athembele kuwo?

Amarobhothi e-AI avame ukuhlanganisa amakhamera, i-LiDAR, izinzwa zokujula, imakrofoni, ama-IMU, ama-encoder, kanye nezinzwa zamandla/i-torque noma ezithintayo. Amamodeli okuqonda aguqula lawa ma-stream abe amasignali asebenzisekayo njengobunikazi bento, ukuma, isikhala esikhululekile, kanye nezinkomba zokunyakaza. Umkhuba omuhle kakhulu osebenzayo ukukhipha ukuzethemba noma ukungaqiniseki, hhayi amalebula nje kuphela. Lokho kungaqiniseki kungaqondisa ukuhlela okuphephile lapho izinzwa ziwohloka ngenxa yokukhanya, ukufiphala, noma ukuxinana.

Iyini i-SLAM kumarobhothi, futhi kungani kubalulekile?

I-SLAM (Ukuhlela Indawo kanye Nokumapha Ngesikhathi Esifanayo) isiza irobhothi ukuthi lakhe imephu ngenkathi lilinganisa indawo yalo ngesikhathi esifanayo. Ibalulekile kumarobhothi ahambahambayo futhi adinga ukuhamba ngaphandle "kokwethuka" lapho izimo zishintsha. Okufakwayo okuvamile kufaka phakathi i-wheel odometry, ama-IMU, kanye ne-LiDAR noma izimpawu zombono, ngezinye izikhathi i-GPS ingaphandle. Ama-stack amahle alandelela ukuzulazula kanye nokungaqiniseki ukuze irobhothi likwazi ukuziphatha ngendlela elondolozayo lapho ukuhlela indawo kuntengantenga.

Ukuhlela amarobhothi nokulawula amarobhothi kuhluke kanjani?

Ukuhlela kunquma ukuthi irobhothi kufanele lenzeni ngokulandelayo, njengokukhetha indawo, ukujikeleza izithiyo, noma ukugwema abantu. Ukulawula kuguqula lolo hlelo lube ukunyakaza okubushelelezi nokuzinzile naphezu kokungqubuzana, izinguquko zomthwalo, kanye nokubambezeleka kwenjini. Ukuhlela kuvame ukuhlukaniswa kube ukuhlela komhlaba wonke (imizila yesithombe esikhulu) kanye nokuhlela kwendawo (ukusabela okusheshayo eduze kwezithiyo). Ukulawula kuvame ukusebenzisa amathuluzi afana ne-PID, ukulawula okusekelwe kumodeli, noma ukulawula kokubikezela kwemodeli ukulandela uhlelo ngokwethembeka.

Amarobhothi abhekana kanjani nokungaqiniseki noma ukuzethemba okuphansi ngokuphephile?

Amarobhothi aklanywe kahle aphatha ukungaqiniseki njengokufaka ekuziphatheni, hhayi into okufanele uyishaye indiva. Lapho ukuqonda noma ukuzethemba kwendawo kwehla, indlela evamile iwukunciphisa ijubane, ukwandisa imingcele yokuphepha, ukuma ngokuphephile, noma ukucela usizo lomuntu esikhundleni sokuqagela. Izinhlelo ziphinde zibhale phansi izenzo kanye nomongo ukuze izehlakalo zihlolwe futhi kube lula ukuzilungisa. Lo mqondo "wokwehluleka okuhle" umehluko omkhulu phakathi kwama-demo namarobhothi angasetshenziswa.

Ukufunda kokuqinisa kuwusizo nini kumarobhothi, futhi yini eyenza kube nzima?

Ukufunda kokuqinisa kuvame ukusetshenziselwa amakhono ayinkimbinkimbi njengokuphatha noma ukuhamba lapho ukuklama isilawuli ngesandla kubuhlungu. Kungathola ukuziphatha okuphumelelayo ngokuzama nokuphutha okuqhutshwa umvuzo, ngokuvamile ekulingiseni. Ukuthunyelwa kuba nzima ngoba ukuhlola kungaba okungaphephile, idatha ingaba yindleko, futhi izikhala ze-sim-to-real zingaphula izinqubomgomo. Amapayipi amaningi asebenzisa i-RL ngokukhetha, kanye nemingcele kanye nokulawula okuvamile kokuphepha nokuzinza.

Ingabe amamodeli ayisisekelo ayashintsha indlela amarobhothi asebenzisa ngayo i-AI?

Izindlela zemodeli yesisekelo zisunduzela amarobhothi ekuziphatheni okuvamile, okulandela imiyalelo, ikakhulukazi ngamamodeli esenzo sombono-ulimi (VLA) njengezinhlelo zesitayela se-RT-2. Okuhle ukuguquguquka: ukuxhumanisa lokho irobhothi elikubonayo nalokho elitshelwe ukuba likwenze nokuthi kufanele lenze kanjani. Iqiniso liwukuthi ukulinganisa okuvamile, imikhawulo yokuphepha, kanye nokulawula okulondoloziwe kusabalulekile ekuthembekeni ngokomzimba. Amaqembu amaningi abheka lokhu njengokuphathwa kwengozi yomjikelezo wokuphila, okufana nomoya nezinhlaka ezifana ne-NIST's AI RMF.

Izinkomba

[1] UDurrant-Whyte noBailey - Ukwenziwa Kwendawo Kanye Nokumamephu Ngesikhathi Esifanayo (SLAM): Ingxenye I Ama-Algorithms Abalulekile (PDF)
[2] ULynch noPark - Amarobhothi Anamuhla: Omakhenikha, Ukuhlela, Nokulawula (I-PDF Ephrintiwe Ngaphambili)
[3] USutton noBarto - Ukufunda Kokuqinisa: Isingeniso (i-PDF yohlelo lwesibili)
[4] I-NIST - Uhlaka Lokuphathwa Kwengozi Yobuhlakani Bokwenziwa (AI RMF 1.0) (PDF)
[5] UBrohan nabanye - RT-2: Amamodeli Ombono-Ulimi-Isenzo Adlulisela Ulwazi Lwewebhu Kulawulo Lwerobhothi (arXiv)

Thola i-AI Yakamuva Esitolo Esisemthethweni Somsizi we-AI

Mayelana NATHI