Lidar-based simultaneous localisation and mapping in marine vehicles: Handling the complex motions in a marine setting
Date
Authors
Type
Examensarbete för masterexamen
Model builders
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
To perform precise and safe manoeuvres in an autonomous vehicle, it is important
to be able to extract its position in relation to its surroundings. Furthermore, an
accurate depth representation of the environment using a lidar scanner helps to
detect distances and possible dangers. Simultaneous localisation and mapping
(SLAM) is a well-researched tool to aid autonomous cars in operation and data
collection to solve both of these tasks at once. The less explored domain of SLAM
on a marine vessel presents more complex kinematic movements and is vulnerable
to drift in the vertical plane. This master thesis is part of the Reeds data set
project, which aims to provide researchers with sensor data to support projects in
marine settings. In this work, it was investigated how to reduce this vertical drift
using hardware and software-based approaches and what the minimal conditions
for performing SLAM are. The baseline performance was built on the lidar-only
NDT and ICP registration algorithms. Data loggings were recorded using a
high-performance lidar with additional sensor data from an inertial measurement
unit and a global navigation satellite system. The latter sensors were fused in an
Unscented Kalman filter to estimate the sensor position and thereby improve
mapping. The hardware-based approach used a gimbal stabiliser to constrain
movement to rotations on the yaw axis. In the analysis, it was shown that both
approaches were able to significantly reduce deviation compared to the lidar-only
approach. While the UKF state estimation performed better in situations with
very sparse point clouds, the gimbal approach offered computational advantages
due to the rotational restrictions. Both approaches can be combined to benefit
from each other. The minimal conditions for SLAM were bound to the point-cloud
density and velocity, as was shown by correlation. These findings can aid in future
data recordings regarding the building of a sensor platform.
Description
Keywords
SLAM, UKF, Gimbal, Mapping, Localisation, Marine systems, Marine datasets, Lidar