INeuroprosthetic entsha yi-AI Robotics Breakthrough

Ososayensi e-EPFL (École polytechnique fédérale de Lausanne) eSwitzerland bamemezele ukuthi kusungulwe uhlobo lokuqala lomhlaba lokulawula ngesandla i-robotic -uhlobo olusha lwe-neuroprosthetic oluhlanganisa ukulawulwa komuntu ne-automation intelligence (AI) ezenzakalelayo ye-robot dexterity enkulu futhi ishicilele ucwaningo lwabo ku Septhemba 2019 ku Ubuhlakani Bomshini Wemvelo .

I-Neuroprosthetics (i-neural prosthetics) yimishini yokufakelwa evuselela noma ithuthukise uhlelo lwezinzwa ngokugqugquzela ugesi ukukhokhela ukusilela okuthinta amakhono emoto, ukuqonda, ukubona, ukuzwa, ukuxhumana, noma amakhono wezinzwa. Izibonelo ze-neuroprosthetics zifaka phakathi i-brain-computer interface (i-BCIs), ukugqugquzela ubuchopho okujulile, izikhuthazi zomgogodla (i-SCS), izimila zokulawula isinye, ukufakelwa kwama-cochlear, kanye neziphutha zenhliziyo.

Inani lamalungu omzimba angenhla lomzimba kulindeleke ukuthi lidlule ku-2.3 billion USD ngo-2025, ngokusho kwezibalo zombiko ka-Agasti 2019 we-Global Market Insight. Ku-2018, inani lemakethe lomhlaba wonke lifinyelele esigidini esisodwa se-USD ngokuya ngombiko ofanayo. Abantu abalinganiselwa ezigidini ezimbili baseMelika banqunywe izitho zomzimba, futhi kunokunqunywa okungaphezu kuka-185,000 okwenziwa minyaka yonke, ngokusho kweNational Limb Loss Information Center. Isifo seVascular sibala amaphesenti angama-82 okunqunywa kwe-U.S. Ngokusho kombiko.

I-myoelectric prosthesis isetshenziselwa ukufaka izitho zomzimba ezinqunyiwe ngomlenze wangaphandle owenziwe osebenza yimisipha yomsebenzisi ekhona. Ngokusho kwethimba labacwaningi le-EPFL, amadivayisi wezohwebo akhona namuhla anganikeza abasebenzisi izinga eliphezulu lokuzimela, kepha ubungqabavu akukho lapho bushesha khona njengobuntu besandla.

“Imishini yezentengiselwano ivame ukusebenzisa uhlelo lokuqopha iziteshi ezimbili ukulawula izinga elilodwa lenkululeko; okusho ukuthi, isiteshi esisodwa se-sEMG sokujika ibhande nesinye sokunweba, ”kubhala abacwaningi be-EPFL ocwaningweni lwabo. “Ngenkathi inembile, uhlelo luhlinzeka ngobuciko obuthile. Abantu balahla izinto zokufakelwa ezikhiqiza amandla kagesi ngamanani aphezulu, ngokwengxenye ngoba benomuzwa wokuthi izinga lokulawula alanele ukufanelekela intengo nobunzima bale mishini. ”

Ukubhekana nenkinga yobungcweti ngama-myoelectric prostheses, abacwaningi be-EPFL bathathe indlela eyahlukene yokwenza lesi sifundo sobufakazi bomqondo ngokuhlanganisa imikhakha yesayensi ye-neuroengineering, robotics, kanye ne-intelligence yokufakelwa ukwenza semi-automate ingxenye yomyalo wezimoto we "shared ukulawula. ”

USilvestro Micera, uSihlalo we-EPFL weBertarelli Foundation ku-Translational Neuroengineering, noProfesa we-Bioelectronics eSchuola Superiore Sant'Anna e-Italy, ubheka le ndlela okwabelwana ngayo yokulawula izandla zerobhothi ingathuthukisa umthelela wasemtholampilo kanye nokusebenziseka kwezinhloso ezahlukahlukene ze-neuroprosthetic ezinjengobuchopho Izindlela zokuhlangana zomshini (BMIs) nezandla ze-bionic.

"Esinye isizathu esenza ukuthi izitho zokufakelwa ezentengiselwano zivame ukusebenzisa ama-classifier-based decoders esikhundleni sezilinganiso ezilinganiselwe kungenxa yokuthi ama-classifiers ahlala ngokuqinile ngendlela ethile," kubhala abacwaningi. “Ukuze ubambe, lolu hlobo lokulawula luhle kakhulu ukuvimbela ukwehla ngengozi kepha ludela i-ejensi yomsebenzisi ngokunciphisa inani lokuma kwesandla okungenzeka. Ukuqaliswa kwethu kokulawulwa okwabiwe kuvumela kokubili i-ejensi yomsebenzisi nokubamba ukuqina. Esikhaleni esikhululekile, umsebenzisi ulawula ngokugcwele ukunyakaza kwesandla, okuvumela futhi ukwakheka kwangaphambili kokuvota kokubamba. ”

Kulolu cwaningo, abacwaningi be-EPFL bagxile ekwakhiweni kwama-software algorithms-i-robotic hardware enikezwe amaqembu angaphandle iqukethe i-Allegro Hand efakwe kurobhothi lwe-KUKA IIWA 7, uhlelo lwekhamera ye-OptiTrack nezinzwa zokucindezela ze-TEKSCAN.

Ososayensi be-EPFL bakhe i-kinematic proportional decoder ngokwakha i-multilayer perceptron (MLP) yokufunda ukuthi ingatolika kanjani inhloso yomsebenzisi ukuze ayihumushe ngokuhamba kweminwe esandleni sokufakelwa. I-multilayer perceptron yinethiwekhi ye-neural yokufakelwa yokuphakela esebenzisa i-backpropagation. I-MLP iyindlela yokufunda ejulile lapho imininingwane iqhubekela phambili ohlangothini olulodwa, kuqhathaniswa nomjikelezo noma iluphu ngenethiwekhi yokufakelwa ye-neural.

I-algorithm iqeqeshwa ngemininingwane yokufaka evela kumsebenzisi owenza uchungechunge lokunyakaza kwesandla. Ngesikhathi sokuhlangana ngokushesha, indlela yeLevenberg – Marquardt yasetshenziselwa ukulinganisa izinsimbi zenethiwekhi esikhundleni sokwehla kwe-gradient. Inqubo yokuqeqeshwa okuyimodeli ephelele ibishesha futhi ithathe imizuzu engaphansi kwengu-10 kusihloko ngasinye, okwenza i-algorithm isebenze ngombono wokusetshenziswa komtholampilo.

"Kumuntu onqunywe isitho, empeleni kunzima kakhulu ukuthola imisipha izindlela eziningi, ezahlukahlukene ukulawula zonke izindlela ezihamba ngeminwe yethu," kusho uKatie Zhuang e-EPFL Translational Neural Engineering Lab, owayengumbhali wokuqala wocwaningo. . “Esikwenzayo ukubeka lezi zinzwa kusiphunzi sazo esisele, bese sizirekhoda bese sizama ukuhumusha ukuthi ziyini izimpawu zokuhamba. Ngoba lezi zimpawu zingaba nomsindo kancane, esikudingayo yile algorithm yokufunda umshini ekhipha umsebenzi onengqondo kusuka kuleyo misipha futhi uyihumushele ekuhambeni. Futhi lokhu kunyakaza yikhona okulawula umunwe ngamunye wezandla zerobhothi. ”

Ngenxa yokuthi ukuqagela komshini kokunyakaza kweminwe kungahle kungabi ngokunembile ngamaphesenti ayi-100, abacwaningi be-EPFL bafaka i-robotic automation ukunika amandla isandla sokuzenzela futhi iqale ngokuzenzekelayo ukuvala into lapho kuthintwa okokuqala. Uma umsebenzisi efuna ukukhipha into, okufanele akwenze nje ukuzama ukuvula isandla ukuze acishe isilawuli se-robotic, abuyisele umsebenzisi kulawulo lwesandla.

Ngokusho kuka-Aude Billard ohola i-Learning Algorithms ne-Systems Laboratory ye-EPFL, isandla se-robotic sikwazi ukusabela ngaphakathi kwama-millisecond angama-400. "Ifakwe izinzwa zengcindezi kuyo yonke iminwe, iyakwazi ukusabela nokusimamisa into ngaphambi kokuba ubuchopho bubone ukuthi into ishibilika," kusho uBillard.

Ngokusebenzisa ubuhlakani bokufakelwa ku-neuroengineering and robotics, ososayensi be-EPFL bakhombisile indlela entsha yokulawulwa okwabiwe phakathi kwenhloso yomshini nomsebenzisi-ukuthuthuka kubuchwepheshe be-neuroprosthetic.

Copyright © 2019 Cami Rosso Wonke Amalungelo Agodliwe.