Push Broom Lidar Licalisation using distance invariant features

This project is an important part of enabling autonomous vehicle navigation in
non-urban environments where it is challenging to obtain position information
from structure. The goal is to precisely localise a vehicle given a 3D map of a
previously traversed trajectory. This task spans in two main categories - teach and repeat phases 4. The
goal at the teach phase is to obtain an accurate prior map from a vertical
(push-broom) and horizontal Laser Rangender (LIDAR) devices mounted on
top of the vehicle 2. Given this map, during the repeat phase the vehicle
should autonomously repeat the path that was initially followed. This is an ill-
dened problem due to the nature of the environment and the different weather
conditions in which the vehicle operates. My approach towards tackling the problem would be to identify distance-
invariant features from the accumulated 3D scans (swathes) created by the
LIDAR in the repeat phase. Comparing these features to the prior map 3
would enable the vehicle to distinguish unique landmarks in the map. Current
approaches for finding such features rely on distance, which is a problem when
traversing similar trajectory multiple times 1. In that case, the LIDAR data
will be deformed due to vehicle rotations and slippage. I will look at different
standard computer vision and machine learning techniques to learn the features
when the state estimate has a high drift rate.
The following are relevant milestones in the project

Obtain teach data, encapsulated in a database
Analise the drawbacks of previous methods and develop an algorithm to
overcome them
Classification of drivable terrain and obstacle avoidance.
Gradually build the aforementioned in a framework for autonomous navigation.
Use the teach data to repeat the behaviour autonomously

References
Ian Baldwin and Paul Newman. Laser-only roadvehicle localization with
dual 2d push-broom lidars and 3d priors. In 2012 IEEE/RSJ International
Conference on Intelligent Robots and Systems, pages 2490{2497. IEEE, 2012.

Ian Baldwin and Paul Newman. Road vehicle localization with 2d push-
broom lidar and 3d priors. In Robotics and automation (ICRA), 2012 IEEE
international conference on, pages 2611{2617. IEEE, 2012.

Nicholas Carlevaris-Bianco, Arash K Ushani, and Ryan M Eustice. Uni-
versity of michigan north campus long-term vision and lidar dataset. The
International Journal of Robotics Research, page 0278364915614638, 2015.

Paul Furgale and Timothy D Barfoot. Visual teach and repeat for long-range
rover autonomy. Journal of Field Robotics, 27(5)534{560, 2010.
2