Reliable Mobile Robot LocalizationPhD

Reliable localization is the problem of determining the position of a mobile with respect to a global or local frame of reference in the presence of sensor noise, uncertainties and potential failures. The basic idea behind many mobile robot localization techniques is to combine sensor data with a priori knowledge about the speci cations of these sensors, the structure of the mobile platform, and the environment the vehicle travels in. For example, it is often assumed that a detailed map of the area is known. In this case, the problem of identifying the position of the robot is the problem of nding an area within the map such that the expected sensor values are at all times in accordance with the actual readings. Many mobile robot applications depend on the ability of the robot to localize using sparse and many times uncertain information concerning its position. For example, fusion of odometric and/or inertial sensors with exteroceptive sensors has received signi cant attention in the past two decades. The techniques mostly used to process sensor data from di erent sources are Kalman ltering and Bayesian estimation. When an initial position estimate is known, a Kalman lter is capable of continuously updating this estimate combining noisy data from a variety of sensors. Bayesian multiple hypothesis testing is suitable for resolving the ambiguity associated with the identi cation of detected landmarks. The proposed study intends to exploit the strengths of both methods. There have been few attempts to combine these two techniques but require strict assumptions. For example it is often assumed that either the motion tracking sensors are perfect or the exteroceptive sensors are capable of extracting a detailed and precise layout of the surroundings. One of the objectives of the proposed research is to introduce a general framework that subsumes both approaches in a single architecture and exploits noisy kinetic information in order to reduce the uncertainty associated with the presence or absence of coarse, sparse, reappearing features of the environment. Restructuring of the exact framework leads to the multiple model adaptive estimation applied here to fault detection and identi cation. It is a common assumption for current localization systems that all sensors provide information at the same rate. Reality dictates that in most cases noisy data from kinetic sensors arrive at high rates while absolute attitude and/or position data tend to be scarce. The intermittent nature of the externally provided or extracted absolute information implies that the system is observable only at certain time instants. This observability limitation restricts the potential autonomy of a robot and causes the loss of precious information whenever real-time ltering is involved. Smoothing of the attitude estimates reduces the overall uncertainty and allows for longer traverses before a new orientation measurement is required. Similar post-processing of the position estimates supports reconstruction of a robot's trajectory and increases the accuracy level during mapping tasks. A simple scenario of an o ce environment with only one type of features is presented to testify the validity of the general approach for combining Bayesian estimation with Kalman ltering in a multiple hypothesis framework. Similarly the simple problem of smoothing the attitude estimates of a robot moving in a 2-D environment is formulated to demonstrate the improvement in the quality of the orientation estimates. It is the objective of this research to ratify the ability of the multiple hypothesis framework to deal with (i) more complicated environments containing di erent types of features with di erent probabilities, (ii) the absence of expected features, (iii) the detection and identi cation of di erent types of failures. In addition, this research will focus on the re-formulation of the smoothing problem in order to be applied in the following cases: (i) smoothing of the attitude estimates in the 3-D case for enhanced localization, (ii) smoothing of the position estimates for enhanced mapping. i Chapter