Abstract:
This research addresses the problem of relative
localization within a robot network possessing relative measurements
between robots. The problem of correspondence is inherent
to most multi-robot relative sensing methods, such as LiDAR,
RADAR and vision based solutions. Multi-sensor multi-target
tracking approaches addresses the problem of correspondence,
when good prioris for initial poses of the sensing platforms are
assumed. However the multi-robot relative localization problem
differs from the classical multi-target tracking scenario due
to; a) the unavailability of initial poses of sensing platforms,
b) the existence of mutual measurements between the sensing
platforms, and c) the measurement set being mixed with both
known and unknown correspondences. To address these specific
characteristics of multi-robot systems, this study proposes a
distributed data correspondence architecture which performs
multi-hypothesis estimation of the robot states. The proposed
architecture is implemented on a multi-robot relative sensor
configuration which possess range measurements with known
data correspondence and bearing measurements with unknown
data correspondence. The proposed distributed multi-robot localization
method is capable of addressing measurement correspondence,
noise, and measurement clutter effectively, while
possessing inherent initialization and recovery capability from
unknown poses.
