Sir Airffic GoʻOku haʻu lahi ʻeni mei he pepa fakatotolo, ka ʻoku tauhi faingofua ʻa e ngaahi lea. Nofo mo au.
Sitepu ʻe fā ʻo e mahino
- ʻIloʻi: kumi ʻa e kakai, meʻalele mo e laine ʻuhila ʻi ha laʻitā pea tā ha puha takatakai ʻiate kinautolu.
- Muimui: muimui ki he feituʻu ʻoku ʻalu ki ai ʻa e meʻa kuo ʻilo.
- ʻIlo hono feituʻu: fakafuofua pe ko e hā hono mamaʻo mo e feituʻu naʻe ngaue ki ai mei he ngaue ʻa e laʻitā mo e IMU (VIO: visual-inertial odometry).
- Mahino mo e fakahinohino: mahinoʻi ʻa e "vakaiʻi ʻa mui ʻi he fale ʻato kulokula" pea fakahoko ʻa e ngaahi lea ki he meʻa ʻoku sio ki ai (VLM: AI ʻoku ne ngāueʻaki fakataha ʻa e laʻitā mo e lea).
Puna lolotonga hono faʻu ʻa e mape: SLAM mo e VIO
ʻI he fakaʻosinga ʻo e 2000 tupu, naʻe kamata ʻe he ngaahi lēpi ʻunivēsiti ke fakafuofua ʻa e feituʻu ʻo ha kiʻi drone ʻaki ha kāmela ʻe taha mo ha IMU. Naʻe kamata ai ʻa e meʻa kotoa.
Naʻe pulusi ʻa e ngaahi founga hangē ko e SVO, VINS-Mono mo e ORB-SLAM pea nau hoko ko e makatuʻunga vahevahe. Naʻe hoko ʻa e fakatotolo ʻatā ko e ngaahi konga ʻo e koloa.
ʻOku lava he ʻaho ni ʻe he drone fakapisinisi ʻo puna ʻi lalo ʻi he halafakakavakava mo loto fale koeʻuhí ko e ngāue ko ʻeni kuo tānaki.
Mei ʻolunga, ʻoku hā iiki mo hehema ʻa e ngaahi meʻa
Ko e AI kuo akoʻi ʻaki ʻa e laʻitā mei he kelekele ʻoku faingataʻaʻia ʻi he sio mei ʻolunga: ʻoku hoko ʻa e kakai ko e kiʻi poini pea liliu ʻa e sīpinga ʻo e meʻalele.
Naʻe faʻu ʻa e ngaahi tānakiʻanga data mei ʻolunga hangē ko e VisDrone mo e DOTA koeʻuhi ke lava ʻo akoʻi ha AI maʻá e langí.
Ko ia ʻoku ʻaʻana ʻa e data ʻoku ne fakafuo ʻa e poto ʻo e AI. ʻOku teʻeki liliu ia ʻi he kuonga ʻo e foundation model.
Sir Airffic GoʻOku faʻa makatuʻunga ʻa e kehekehe ʻo e AI ʻi he lahi ʻo e data mei ʻolunga ʻoku ke maʻu, lahi ange ia ʻi he ngaahi founga fakakomipiuta. Manatuʻi ʻeni ki he vahe palani.
Ko e ʻaho naʻe ikuna ai ʻe he AI ʻa e kau tauhi pule ko e tangata
ʻI he 2023, naʻe ikuna ʻe ha drone AI ko Swift, naʻe ngaohi ʻi he ʻUnivēsiti ʻo Zurich, ʻa e kau tauhi pule ʻa e māmani ʻi he lova drone. Naʻe pulusi ʻa e ngāue ʻi he Nature.
Naʻe fakafuofua ʻe Swift hono feituʻu ʻaki pē ha kāmela mo ha IMU, pea naʻe ako ke puna ʻaki ʻa e reinforcement learning — ko e ako ʻaki hono toutou feinga ʻi ha simulation.
Naʻe fai ʻeni ʻi ha hala lova kuo fokotuʻu. Ko hono aʻusia ʻa e poto ko ia ʻi he langi moʻoni, mo e liliu ʻa e ʻea mo e meʻa fakataʻotaʻofa, ʻoku kei pole pē ia.
Lova drone ʻi he tuʻunga tauhi pule ʻaki ʻa e deep reinforcement learning (UZH)
Fakamoʻoni: [N124]
Ko e peau hoko: sio, fakakaukau mo ngāue ki he lea
Mei he 2023 nai, naʻe kamata ke hoko ʻa e ngaahi mōtolo AI lalahi ʻoku nau ngāueʻaki fakataha ʻa e laʻitā mo e lea (foundation model) ko e ʻuto ʻo e lopote.
ʻOku tupulaki ʻa e fakatotolo drone ki he muimui ki he fakahinohino ʻi he lea lolotonga hono mahinoʻi ʻa e ʻātakai (aerial vision-language navigation).
ʻOku mamafa ʻa e ngaahi mōtolo lalahi, pea ʻoku kei faingataʻa ke fakalele vave ʻi he vakapuna. Ko ia ʻoku pau ai ke lau fakataha mo e ngaahi chip — ko e vahe hoko.
Fakamoʻoni: [N133]
Ngaahi fehuʻi ʻoku faʻa ʻeke
Ko e fē ngaahi ngāue ʻoku ngāueʻaki ai ʻa e sio ʻa e komipiuta ʻa e drone?
Ko e kumi ʻo e mafahifahi ʻi he halafakakavakava mo e laine ʻuhila, siviʻi ʻo e moʻui lelei ʻa e ngoue, kumi kakai ʻi he kumi mo fakahaofi, leʻohi mo e vakaiʻi ʻo e fefonongaʻaki, mo e siviʻi ʻo e feituʻu ʻave koloa ko e ngāueʻanga angamaheni ia.
Ko e hā ʻa e visual-inertial odometry (VIO)?
Ko ha founga ʻoku ne fakatahaʻi ʻa e ngaue ʻa e laʻitā mo e fua ʻa e IMU ʻo e vave mo e takai ke fika ʻa e mamaʻo naʻe ngaue ai ʻa e drone. Ko ha tekinolosia tefito ia ki he puna taʻe-GPS.