Sir Airffic GoLokunyenti lapha kuvela emaphepheni elucwaningo, kodvwa emagama alula. Hamba nami.
Tigaba letine tekucondza
- Kutfola (detection): kutfola bantfu, timoto netintsambo tagesi esitfombeni bese udvweba emabhokisi lativungeletako.
- Kulandzela (tracking): kulandzela kutsi intfo letfoliwe iya kuphi ngemuva kwaloko.
- Kutati indzawo: kubala kutsi uhambe kangakanani nekutsi uye kuphi ngekunyakaza kwesitfombe kanye ne-IMU (VIO: visual-inertial odometry).
- Kucondza nekulalela ticondziso: kucondza “hlola ngemuva kwendlu lenophahla lolubovu” bese uxhumanisa emagama naloko lokubonako (VLM: i-AI lephatsa titfombe nelulwimi ndzawonye).
Kundiza ngesikhatsi wakha libalave: i-SLAM ne-VIO
Ngasekupheleni kweminyaka ya-2000, emalabhorethri emanyuvesi acala kubala indzawo ye-drone lencane ngekhamera yinye ne-IMU. Konkhe kwacala lapho.
Tindlela letinjenge-SVO, VINS-Mono ne-ORB-SLAM tashicilelwa futsi taba sisekelo lesabelwanako. Lucwaningo lolusobala lwaphenduka tincenye temikhicito.
Ema-drone emsebenti anamuhla angandiza ngaphansi kwemabhuloho nangekhatsi etakhiweni ngenca yalomsebenti lowakheka ngekuhamba kwesikhatsi.
Kusuka etulu, tintfo tibukeka tincane futsi tigobekile
I-AI lefundziswe ngetitfombe letitsatfwe emhlabatsini iyahlupheka nayibuka kusuka etulu: bantfu baba ngemachashazi, timoto tishintja simo.
Ema-dataset asemoyeni njenge-VisDrone ne-DOTA akhiwa kuze kufundziswe i-AI lelungele sibhakabhaka.
Loyo lophetse idatha uncuma kutsi i-AI ihlakaniphe kangakanani. Loku akukashintji nasesikhatsini sema-foundation model.
Sir Airffic GoEmagebe e-AI avame kuvela ebunyentini bedatha yasemoyeni lonayo, kakhulu kunema-algorithm. Khumbula loku esehlukweni semasu.
Lilanga i-AI lehlula ngalo bompetsa babantfu
Nga-2023, i-drone ye-AI lebitwa ngekutsi yi-Swift, leyakhiwa eNyuvesi yase-Zurich, yehlula bompetsa bemhlaba babantfu bekugijimisa ema-drone. Lomsebenti washicilelwa ku-Nature.
I-Swift yabala indzawo yayo ngekhamera ne-IMU kuphela, futsi yafundza kundiza nge-reinforcement learning — kutijwayeta kanyenti ku-simulation.
Loku bekusendleleni lengashintji. Kufinyelela lelikhono esibhakabhakeni sangempela, lapho simo selitulu netiphazamiso kushintja khona, kuseyinselele.
Kugijimisa ema-drone ezingeni lebompetsa nge-deep reinforcement learning (UZH)
Bufakazi: [N124]
Ligagasi lelilandzelako: kubona, kucabanga nekwenta ngemagama
Kusukela nga-2023, ema-model e-AI lamakhulu laphatsa titfombe nelulwimi ndzawonye (ema-foundation model) acala kusebenta njengengcondvo yemarobhothi.
Lucwaningo lwe-drone ngekulandzela ticondziso letikhulunywako ngesikhatsi icondza lokuyizungelezile (aerial vision-language navigation) luyakhula.
Ema-model lamakhulu ayasindza, futsi kuwasebentisa ngekushesha endizeni kusamatima. Ngiko nje ema-chip — sehluko lesilandzelako — kufanele afundvwe kanye nalokhu.
Bufakazi: [N133]
Imibuto levame kubutwa
Ngumiphi imisebenti lesebentisa kubona kwekhompyutha kwe-drone?
Kutfola imifantfo emabhulohweni nasetintsanjeni tagesi, kuhlola imphilo yetitjalo, kufuna bantfu ekusindziseni, kuphepha nekubuka timoto, nekuhlola tindzawo tekuletsa timphahla kuyindlela leyejwayelekile.
Yini i-visual-inertial odometry (VIO)?
Yindlela lehlanganisa kunyakaza kwesitfombe netilinganiso te-IMU tekusheshisa nekujika kuze ibale kutsi i-drone ihambe kangakanani. Bubuchwepheshe lobusisekelo bekundiza ngaphandle kwe-GPS.