Cooperative Localization: On motion-induced initialization and joint state estimation under communication constraints.

Abstract

Teams of mobile robots are becoming increasingly popular to measure and estimate quantities of interest at spatially distributed locations. They have been used in tasks such as surveillance, search and rescue, and underwater- or space exploration. For these tasks, accurate localization, i.e., determining the position and orientation (pose) of each robot, is a fundamental requirement. Instead of localizing each robot in a team independently, Cooperative Localization (CL) incorporates robot-to-robot observations and jointly estimates all robots' poses, which improves localization accuracy for all team members. However, such joint estimation also creates significant challenges. In particular, initializing a joint estimation algorithm requires knowledge of all robots' poses with respect to a common frame of reference. This initialization is straightforward using GPS or manual measurements, but is difficult in the absence of external references. The second difficulty of CL is that it requires communicating large amounts of data, e.g., the robots' sensor measurements or state estimates. However, transmitting all these quantities is not always feasible, either due to bandwidth or power constraints. This thesis offers novel solutions to the aforementioned problems. In the first part of the thesis, we investigate the problem of CL initialization, using robot-to-robot measurements acquired at different vantage points during robot motion. We focus on the most challenging case of distance-only measurements, and provide algorithms that compute the guaranteed global optimum of a nonlinear weighted Least Squares problem formulation. These techniques exploit recent advances in numeric algebraic geometry and optimization. In the second part, we investigate the problem of CL under communication constraints. To reduce communication bandwidth, we propose using adaptively quantized measurements. We extend existing quantized filtering approaches to batch MAP estimators, and apply these techniques to multi-robot localization. We provide results on optimal threshold selection, as well as optimal bit allocation to efficiently utilize time-varying bandwidth. Our results are validated in simulation and experiments. By providing solutions for two important problems in CL { motion-induced estimator initialization, and estimation under communication constraints { the research presented in this thesis aims to promote use of cooperative mobile robots in challenging real-world applications.