Rao-Blackwellized Particle Filters SLAM with sidescan sonar
Title: Rao-Blackwellized Particle Filters SLAM with sidescan sonar
SNIC Project: Berzelius-2022-128
Project Type: LiU Berzelius
Principal Investigator: Yiping Xie <yipingx@kth.se>
Affiliation: Kungliga Tekniska högskolan
Duration: 2022-06-03 – 2023-01-01
Classification: 10207
Homepage: https://pytorch3d.org/
Keywords:

Abstract

Differentiable rendering has been applied to computer graphics recently, combining deep learning, especially neural representation. The advantage of it is that there is no need to acquire 3D annotation, which is expensive. One can only have 2D annotations on images and leverage differentiable rendering to compute loss in 2D camera images and back-propagate the gradients back to whatever is needed to optimize. We have a similar situation in underwater perception, especially on imaging sonars such as sidescan sonars. It is expensive to register sidescan sonar data to a 3D seafloor map, which is no longer needed if we can utilize the idea of differentiable rendering. There is an open-source software pytorch3d (https://pytorch3d.org/) where we can use and modify the camera model to a sonar model and do 3D reconstruction, localization and SLAM with sidescan sonar only. IN particular, with the tool of differentiable rendering it is possible to use Rao-Blackwellized Particle Filters to do SLAM with sidescan sonar data. The training and optimization would need many GPU resources, hence the SNIC computing resources would be of great help of the said project. And the project would help autonomous underwater vehicles increase aotounomy.