Thesis title: Enhancing Spatio-Temporal Scalability in Graph-based SLAM systems
In robotics, Simultaneous Localization and Mapping (SLAM) is the fundamental
task of estimating the location of an autonomous agent while building a digital
model of the environment in which the agent operates. Its importance is critical
for autonomy as, to complete any physical task in the real world, the agent needs
to know where it is and how the world is structured. Moreover, the digital representation of
the world needs to contain a sufficient amount of geometric and semantic details to support
the (potentially) large variety of tasks the agent needs to complete. For example, a mobile
robot operating in a hospital needs to know its location in the building, the floor’s structure,
the location of objects of interest (e.g., drugs, tools etc.), and the patient’s identity in each
room to support nurses and doctors in the healthcare task.
Although mobile robots are rapidly becoming an integral part of our everyday life,
their applications in the real world are usually limited to short-term missions in relatively
small and controlled environments. The leading causes are the current hardware and
algorithmic limitations that do not allow modern autonomous robots to deal with large-scale
environments for extended periods.
In this thesis, we investigate how to extend the current capabilities of modern SLAM
systems in terms of the size of the environment and the timeframe in which they can operate
robustly. More specifically, we will revise the standard modules that compose a SLAM
pipeline to tackle the algorithmic limitations that prevent them from scaling up in terms of
space they can explore and the amount of time they can reliably operate.
Most of the contributions presented in this thesis are available as open-source software
packages. We believe that this will foster collaboration, enable precise repeatability of our
experiments, and facilitate future research on Spatio-temporal scalability in SLAM.