Slawomir Grzonka
Mapping, State Estimation, and Navigation for Quadrotors and Human-Worn Sensor Systems
Technical support for first responders, who have to explore hazardous environments to locate
and rescue victims and thereby avoiding dangerous areas, has recently gained a substantial interest
in the research community.
One possibility to assist first responders is by using robots to explore the environment and deitect
dangerous areas or locate victims. Flying robots are envisioned to play a key role in this context
as their increased mobility allows them to fly over obstacles where wheeled robots get stuck. In
order to be able to operate indoors, the flying robot should be able to keep a stationary pose. All
these prerequisites are met by quadrotors. Unfortunately, teaching human personnel to remotely
steer such a flying platform is time intensive and costly. Additionally, manual piloting bears the
risk of damaging the robot due to the confined space indoors and due to environmental conditions
that can have a severe impact on the quality of the radio link. The flying robot therefore
needs to be able to operate autonomously over an extended period of time. In this case, only
minimal input from a human (e.g., the next location the robot should fly to) is required.
Another possibility to support first responders is by using sensor systems that are integrated
into their garment. Such sensors can provide vital information about the current location of the
wearer or an approximate map of the environment. In the context of first responders operating
in an unknown building, this information could be used to re-route the wearer to the nearest exit
in case of emergency, especially if environmental conditions like smoke and fire elicit confusion
among first responders. Even more, this kind of information can be employed in a search and
rescue mission by improving the delegation of different teams, i.e., by avoiding searching the
same area multiple times.
However, systems like the ones described so far (i.e., sensor systems mounted on a quadrotor or
integrated into the garment) need to be aware of their current state, including their own location.
To estimate the location indoors, a map of the environment is needed in most cases. In general,
this map is not known beforehand and the (robotic) system needs to build a map of the environment
based on its sensor measurement during the mission. Due to the limited computational
power available, efficient mapping techniques are mandatory.
In this thesis we develop novel technologies for efficient mapping that can be used with a variety
of sensors. We furthermore develop a navigation system that enables a small-sized flying
robot (quadrotor) to fly autonomously indoors. Finally, we develop an approach to map indoor
environments based on human motion recorded with a data suit, i.e., an embedded sensor system
consisting of several inertial measurement units worn by the human.
In the first part of this thesis, we present an innovative technique for estimating the trajectory of
a robot, given its observations. We will demonstrate that compared to other state-of-the-art approaches
our approach is up to several orders of magnitude faster without any loss in accuracy.
This allows embedded systems to efficiently and accurately recover their trajectory and thus allows
them to build accurate maps of the environment. This general framework is subsequently
used in the second part of the thesis in the corresponding embedded sensor systems.
In the second part, we develop two navigation systems for different types of sensor setups. The
first navigation system enables a small-sized quadrotor to fly autonomously indoors. This includes
pose control, localization, map building, path-planning, and obstacle avoidance. We also
present a novel technique to map obstacles underneath the robot. The second system employs
information recorded with a data suit consisting of several inertial measurement units worn by a
human. We develop a solution to recover the trajectory of the human and to build a geometrical
as well as topological map of the environment. In all cases, we solely employ the motions and
detected activities of the human. The wearer, therefore, does not need to carry any additional
sensors like cameras or laser scanners that would also not work reliably in the case of environmental
conditions like smoke and fire.
Above, we motivated our work in particular envisioned for first responders. However, it is
important to note that the developed technologies can be applied to a variety of scenarios and
are not restricted to the field of search and rescue.
Bibtex:
@phdthesis{grzonkaPhd,
author = {Grzonka, Slawomir},
title = {Mapping, State Estimation, and Navigation for Quadrotors and Human-Worn Sensor Systems},
school = {Albert-Ludwigs-University of Freiburg},
year = {2011}
}
Document (pdf file):
thesis (41MB)