Abstract:
Unmodeled systematic and nonsystematic errors in robot
kinematics and measurement processes often cause adverse effects in
several autonomous navigation tasks. In particular, accumulated
sensor biases can render simultaneous localization and mapping
(SLAM) algorithms of autonomous vehicles to perform very poorly
especially in large unexplored terrains including cycles, as a result of
the estimator divergence and inconsistency. One way to deal with
this problem is the accurate modeling and precise calibration of
sensors. However this may add up to longer setup and calibration
times. Even after accurate calibration and modeling, sensor
calibration may often subject to drifts, rendering the efforts
ineffective. Therefore, the correct and effective way to deal with this
problem is explicit estimation of these parameters with other states.
In this work we address the estimation theoretic sensor bias
correction problem in SLAM using a simple unified framework and
establish theoretically, the behavior and properties of the solution
with special consideration to diminishing uncertainty, rates of
convergence and observability.
