Cooperative Localization and Multi-Robot Exploration

Abstract

This thesis has two main contributions. The first contribution is the
use  of cooperative localization for  decoupling  the positional error
of a moving robot from its environment.  The second contribution is
the development of efficient multi-robot exploration strategies for an
unknown environment.

The proposed method is designed to be robust in the face of
arbitrarily large odometry errors or objects with poor reflectance
characteristics. Central to  the exploration strategy is a  sensor
(robot tracker) mounted on a robot that could track a second mobile
robot and accurately report its relative position. Our exploration
strategies use the robot tracker sensor to sweep areas of free space
between stationary and moving robots and to generate a graph-based
description of the environment. This graph is used to guide the
exploration process. Depending on the size of the environment relative
to the range of the robot tracker, different spatial decompositions
are used: a triangulation or a trapezoidal decomposition of the free
space. Complete exploration without any overlaps is guaranteed as a
result of the guidance provided by the dual graph of the spatial
decomposition of the environment.

The uncertainty in absolute robot positions and the resulting
uncertainty in the map is reduced through the use of a probabilistic
framework based on particle filtering (a Monte Carlo simulation
technique). Particle filtering is a probabilistic sampling technique
used to efficiently  model complex probability distributions that
cannot be effectively described using classical methods (such as
Kalman filters).

We present experimental results from two different implementations of
the robot tracker sensor, in simulated and in real environments. The
accuracy of the resulting map increases with the use of cooperative
localization. Furthermore, the deterioration of the floor conditions
did not affect the quality of the map verifying the decoupling of
positioning error from the environment.