Airffic InstituteAERIAL PHYSICAL AI
Zvirimo ☰

Aerial Physical AI Institute

Vision AI: kuti drone inonzwisisa sei zvainoona

Kuona hakuna kukwana kuti ubhururuke. Uropi hunoda kunzwisisa zvahunoona.

Mhinduro pfupi: Vision AI yemadrone yakakura nematanho mana: ① kuwana zvinhu (detection), ② kuzvitevera (tracking), ③ kuziva pauri (SLAM, VIO), ④ kunzwisisa mirairo yemashoko (foundation models). Tiri pamusuwo wedanho ④.

Yakavandudzwa: · Maminitsi 3 ekuverengaTakatsvagurudza sei →

Sir Airffic GoZvizhinji pano zvinobva mumapepa etsvagurudzo, asi mashoko akachengetwa ari nyore. Ramba uneni.

Matanho mana ekunzwisisa

  1. Kuwana zvinhu (detection): kuwana vanhu, mota netambo dzemagetsi mumufananidzo nekuzvidhirowera mabhokisi.
  2. Kutevera (tracking): kutevera kuti chinhu chawanikwa chiri kuenda kupi.
  3. Kuziva pauri: kuverenga kuti wafamba zvakadii nekupi uchishandisa kufamba kwemufananidzo neIMU (VIO: visual-inertial odometry).
  4. Kunzwisisa nemirairo: kubata kuti "tarisa kuseri kweimba ine denga dzvuku" nekubatanidza mashoko nezvaunoona (VLM: AI inobata mifananidzo nemutauro pamwe chete).

Kubhururuka uchiita mepu: SLAM neVIO

Mukupera kwema2000, marabhu emayunivhesiti akatanga kuverenga nzvimbo yedrone diki nekamera imwe neIMU. Zvose zvakatangira ipapo.

Nzira dzakaita seSVO, VINS-Mono neORB-SLAM dzakaburitswa dzikava hwaro hunogovaniswa. Tsvagurudzo yakavhurika yakava zvikamu zvezvigadzirwa.

Madrone ebhizinesi emazuva ano anogona kubhururuka pasi pemabhiriji nemukati medzimba nekuda kwebasa iri rakaunganidzwa.

Humbowo: [P01] [P02] [P08] [P09] [P21] [P15]

Kubva kumusoro, zvinhu zvinoita sezviduku uye zvakaminama

AI yakadzidziswa nemifananidzo yakatorwa pasi inonetseka nemaonero ekumusoro: vanhu vanova madotsi uye mota dzinochinja chimiro.

Madataset emudenga akaita seVisDrone neDOTA akavakwa kuti AI yemudenga idzidziswe.

Uyo ane data ndiye anosarudza kungwara kweAI. Izvi hazvina kuchinja munguva yemamodheru makuru.

Sir Airffic GoMisiyano yeAI kazhinji inobva pahuwandu hwedata remudenga raunaro, kupfuura maalgorithm. Rangarira izvi pachitsauko chezano.

Humbowo: [P14] [P23] [P24]

Zuva rakakunda AI vashasha vevanhu

Muna 2023 drone yeAI inonzi Swift, yakagadzirwa paUniversity of Zurich, yakakunda vanhu vashasha vepasi rose vemijaho yemadrone. Basa iri rakaburitswa mujenari reNature.

Swift yaiverenga nzvimbo yayo nekamera neIMU chete, uye yakadzidza kubhururuka kuburikidza nereinforcement learning — ichidzidzira kakawanda musimulator.

Izvi zvaiva panzira yakatarwa. Kuwana unyanzvi uhwu mudenga chairo, mamiriro ekunze nezvipingamupinyi zvichichinja, kuchiri dambudziko risati ragadziriswa.

Mujaho wemadrone wepamusoro-soro uchishandisa deep reinforcement learning (UZH)

Vhidhiyo yedrone yeAI Swift yakaburitswa neRobotics and Perception Group yeUniversity of Zurich. YouTube ↗ · Hapana chinotorwa kubva kuYouTube kusvikira wadzvanya tamba

Humbowo: [N124]

Mafungu anotevera: kuona, kufunga nekuita zvichibva pamashoko

Kubva kunenge 2023, mamodheru makuru eAI anobata mifananidzo nemutauro pamwe chete (foundation models) akatanga kushanda seuropi hwemarobhoti.

Tsvagurudzo yemadrone anotevera mirairo inotaurwa achinzwisisa zvaari kuona (aerial vision-language navigation) iri kukura.

Mamodheru makuru anorema, uye kuamhanyisa nekukurumidza mundege kuchiri kwakaoma. Ndosaka machipi — chitsauko chinotevera — anofanira kuverengwa pamwe chete.

Humbowo: [N133]

Mibvunzo inowanzobvunzwa

Mabasa api anoshandisa computer vision yemadrone?

Kuwana mitswe pamabhiriji netambo dzemagetsi, kuongorora hutano hwezvirimwa, kutsvaga vanhu pakununura, kuchengetedza nekutarisa traffic, uye kuongorora nzvimbo dzekuendesa zvinhu ndizvo zvinowanzoitwa.

Chii chinonzi visual-inertial odometry (VIO)?

Inzira inosanganisa kufamba kwemufananidzo nezvinoyerwa neIMU zvekumhanya nekutenderera kuti iverenge kuti drone yafamba zvakadii. Ihunyanzvi hwekutanga hwekubhururuka pasina GPS.