PhD Thesis

Information Metrics for Localization and Mapping

Work default illustration


  • Started: 01/01/2014
  • Finished: 27/02/2019


Decades of research have made possible the existence of several autonomous systems that successfully and efficiently navigate within a variety of environments under certain conditions. One core technology that has allowed this is simultaneous localization and mapping (SLAM), the process of building a representation of the environment while localizing the robot in it.
State-of-the-art solutions to the SLAM problem still rely, however, on heuristic decisions and options set by the user. In this thesis we search for principled solutions to various aspects of the localization and mapping problem with the help of information metrics.
One such aspect is the issue of scalability. In SLAM, the problem size grows indefinitely as the experiment goes by, increasing computational resource demands. To maintain the problem tractable, we develop methods to build an approximation to the original network of constraints of the SLAM problem by reducing its size while maintaining its sparsity. In this thesis we propose three methods to build the topology of such approximated network, and two methods to perform the approximation itself.
In addition, SLAM is a passive application. It means, it does not drive the robot. The problem of driving the robot with the aim of both accurately localizing the robot and mapping the environment is called active SLAM. In this problem two normally opposite forces drive the robot, one to new places discovering unknown regions and another to revisit previous configurations to improve localization. As opposed to heuristics, in this thesis we pose the problem as the joint minimization of both map and trajectory estimation uncertainties, and present four different active SLAM approaches based on entropy-reduction formulation.
All methods presented in this thesis have been rigorously validated in both synthetic and real datasets.

The work is under the scope of the following projects:

  • Cargo-ANTS: Cargo handling by Automated Next generation Transportation Systems for ports and terminals (web)
  • LOGIMATIC: Tight integration of EGNSS and on-board sensors for port vehicle automation (web)