Пропустить навигацию
Testing Microduck's SLAM!
0:00 / 1:00

Testing Microduck's SLAM!

Antoine Pirrone

@antoinepirrone5 тыс. подписчиков

Открыть в X
238 отметок «Нравится»СкачатьОригинальный пост

14 367 просмотров

Testing Microduck's SLAM! It runs fully onboard, with just its leg odometry and the 8×8 ToF sensor in its head, 64 pixels of depth. Carry it across the room and it finds itself again after a few steps. The ArUco tags are only there for ground truth: a fixed camera tracks the one on its head, and the duck never uses them. Against that ground truth, its position at each stop is off by 4.8 cm on median, and 94% of the time it claims to know where it is, it's within 15 cm. In this session it relocalized after a 1.5 m carry (2 cm off), a seated 1.5 m carry (11 cm) and a fall (3 cm), and across our benchmark sessions it never relocalized to the wrong place.

10 комментариев

Комментарии из оригинального поста в X (Twitter)

  • Фото профиля Quentin Munch

    @Neurotronic67

    It remind me of TinySLAM : Steux, B., & El Hamzaoui, O. (2010, December). tinySLAM: A SLAM algorithm in less than 200 lines C-language program. (pp. 1975-1979). IEEE. the code is so small it's in the paper !

    1 отметка «Нравится»

  • Фото профиля JeCo

    @Je__Co

    That's impressive for this little duck!

  • Фото профиля TaraT

    @tarat_211

    the aruco tags give me headacheeeeee

  • Фото профиля ζ Pedram ζ

    @zenstyle

    Cool duckwork, Looked at optical flow?

  • Фото профиля 01azPublic

    @silic2023

    How much of a run does it spend in TRACKING versus trying to relocalize? That would help read the 94%-within-15-cm result: both the accuracy when confident and how often that confidence is available matter for navigation.

  • Фото профиля satvik

    @satvikpathak007

    the seated carry is the nasty test here. zero steps, 1.5 m of displacement. the map has to win that argument.

  • Фото профиля AetrixEmby

    @AetrixEmby

    只靠腿部里程计和8x8 ToF就能全机跑SLAM,这集成度有点厉害

  • Фото профиля Alvaro L

    @L42ARO

    Is if Monocular slam?

  • Фото профиля han

    @han667807560546

    This is seriously impressive! Achieving SLAM with just leg odometry and an 8×8 ToF sensor is amazing. Curious how you handle relocalization with such limited depth data. Great work! 🔥

  • Фото профиля Aayushman Singh

    @aayushman2703

    This is the kidnapped robot problem on nightmare difficulty lol