Sir Airffic GoPlante samting long ia i kamaot long ol pepa blong risej, be mi kipim ol tok i isi. Stap wetem mi.
Fo step blong andastanem
- Faenem: faenem ol man, trak mo paoa laen long wan pikja mo droem bokis raon long olgeta.
- Folem: folem wea samting we yu faenem i go nekis.
- Save ples blong yu: faenemaot hamas mo wea yu muv, long muv blong pikja wetem IMU (VIO: visual-inertial odometry).
- Andastanem mo folem oda: andastanem “jekem biaen long haos we ruf blong hem i red” mo joenem ol tok wetem samting we yu luk (VLM: AI we i wok wetem pikja mo tok tugeta).
Flae mo mekem map: SLAM mo VIO
Long en blong ol yia 2000, ol lab blong yunivesiti oli stat blong faenemaot ples blong wan smol drone wetem wan kamera mo wan IMU. Evri samting i stat long ples ia.
Ol rod olsem SVO, VINS-Mono mo ORB-SLAM oli kamaot mo oli kam fandesen we evriwan i serem. Risej we i open i kam ol pat blong ol prodak.
Tede, ol drone blong bisnis oli save flae andanit long brij mo insaed long haos from ol wok ia we i hip i kam antap.
Long antap, ol samting oli luk smol mo oli no stret
AI we oli tijim wetem ol foto long graon i gat trabol wetem ol pikja long antap: ol man oli kam olsem smol dot mo fom blong ol trak i jenis.
Ol data blong pikja long skae olsem VisDrone mo DOTA oli mekem blong tijim AI we i stret blong skae.
Hu we i holem data i disaedem hamas AI i kam smat. Hemia i no jenis long taem blong foundation model.
Sir Airffic GoDefren bitwin ol AI i kamaot plante taem long hamas data blong skae yu holem, moa long ol algorithm. Tingbaot samting ia blong japta blong strateji.
Dei we AI i winim ol nambawan man
Long 2023, wan AI drone we nem blong hem Swift, we oli mekem long Yunivesiti blong Zurich, i winim ol nambawan man blong wol long resis blong drone. Wok ia i kamaot long Nature.
Swift i faenemaot ples blong hem wetem kamera mo IMU nomo, mo i lanem blong flae tru long reinforcement learning — i traem plante taem insaed long simulator.
Hemia i long wan rod blong resis we i no jenis. Blong mekem semak save long skae blong tru, wetem weta mo ol samting we i jenis, hem i wan had wok yet.
Drone resis we i kasem leva blong nambawan, wetem deep reinforcement learning (UZH)
Pruf: [N124]
Nekis wef: luk, tingting mo mekem wanem tok i talem
Stat klosap long 2023, ol bigfala AI model we i wok wetem pikja mo tok tugeta (foundation model) oli stat blong kam brein blong ol robot.
Risej long drone we i folem oda long tok taem hem i andastanem ples (aerial vision-language navigation) i stap gro.
Ol bigfala model oli hevi, mo blong ranem olgeta kwik insaed long drone i had yet. Hemia nao from wanem yu mas ridim ol chip — nekis japta — wetem.
Pruf: [N133]
Ol kwestin we oli askem plante
Wanem ol wok we oli yusum kamera AI blong drone?
Faenem ol brok long brij mo paoa laen, jekem helt blong ol garen, lukaotem man blong sevem olgeta, sekiuriti mo lukaot long trafik, mo jekem ples blong karem samting, hemia ol wok we oli yusum moa.
Wanem ia visual-inertial odometry (VIO)?
Wan rod we i joenem muv blong pikja wetem mesa blong IMU blong spid mo raon, blong kaontem hamas drone i muv. Hem i wan teknoloji we i stamba blong flae we i no gat GPS.