Plane SLAM: SLAM backend benchmarking

Plane SLAM: about whole project

Planes are basic primitives in man-made environments that could be extracted from 3D data (LiDARs, depth cameras). Those primitives can serve as landmarks in SLAM for localization and mapping of autonomous systems (self-driving cars, indoor robots – service robots, warehouse robots). Plane SLAM project is devoted to research on these primitives in two directions: (1) develop methods for plane extraction from heterogeneous sensors and (2) elaborate on how those primitives could be used in SLAM backend.

To get intuition on SLAM and its application in robotics: https://ieeexplore.ieee.org/document/7747236 About this topic in Russian: https://drive.google.com/file/d/15Mn4ZdC4TOcF0fuU1EFhdj9VYEKhgScK/view?usp=sharing

Current state: we have a benchmark of existing plane extraction approaches for depth camera and LiDAR, that includes labeled dataset, metrics, performance evaluation of existing methods for plane extraction.

All topics related to this project could be considered as an opportunity to be on board for publications in top-level robotics conferences and journals (in case of generating smth useful for project, ofc).

Key members of the project from SPBU side: Dmitrii Iarosh (M.Sc., 1st year), Pavel Mokeev (B.Sc., 2nd year)

Task. Planar SLAM benchmark

The task is to evaluate performance of existing planar SLAM backends on our dataset. That includes the next subtasks:

  1. Run existing methods of Planar SLAM (check links below)
  2. Adopt methods to our dataset format
  3. Dockerize methods to make experiments reproducible
  4. Evaluate performance using standard metrics

Planar SLAM backends to be evaluated (not limited to this list)

Technologies used in project: Python, C++, Docker, Bash, CI/CD (GA)

Требования к студенту

Experience with any mentioned technologies will be a great bonus, but we don't expect that students should have a strong background in them, the most important thing is to be open to work intensely and hard, don't be scared a lot from math.

Уровень

2 курс, 3 курс, Бакалаврская ВКР


Руководитель

Литвинов Юрий Викторович


Консультант

Kornilova Anastasiia Валерьевна


Источник

Mobile Robotics Lab, Skoltech