Robust mobile robot localization: From single-robot uncertainties to multi-robot interdependencies

Robust localization is the problem of determining the position of a mobile robot with respect to a global or local frame of reference in the presence of sensor noise, uncertainties and potential failures. Previous work in this field has used Kalman filters to reduce the effects of sensor noise on updates of the vehicle position estimate or Bayesian multiple hypothesis to resolve the ambiguity associated with the identification of detected landmarks. This dissertation introduces a general framework for localization that subsumes both approaches in a single architecture and applies it to the general problem of localizing a mobile robot within a known environment. Odometric and/or inertial sensors are fused with data obtained from exteroceptive sensors. The approach is validated by solution of the “kidnapped robot” problem.; The second problem treated in this dissertation concerns the common assumption that all sensors provide information at the same rate. This assumption is relaxed by allowing high rate noisy odometric or inertial data from kinetic sensors while absolute attitude and/or position data (e.g., from sun sensors) are obtained infrequently. We address the resulting observability limitation by incorporating a Smoother in the attitude estimation algorithm. Smoothing of the attitude estimates reduces the overall uncertainty and allows for longer traverses before a new absolute orientation measurement is required. Simulation examples also show the ability of this method to increase the accuracy of robot mapping.; The third problem concerns multiple robots collaborating on a single task. In prior research with a group of, say M, robots the group localization problem is usually approached by independently solving M pose estimation problems. When collaboration among robots exists, current methods usually require that at least one member of the group holds a fixed position while visual contact with all the other members of the team is maintained. If these two conditions are not met, uncorrelated pose estimates can lead to overly optimistic estimates. We introduce a new distributed Kalman filtering approach for collective localization that overcomes the previous limitations and combines optimally all available positioning information amongst the group members.