Simultaneous stochastic mapping and localization

Abstract

In order to create truly autonomous mobile robots, the task of building an accurate map of an a priori unknown environment and concurrently using that map to navigate is a central problem. This thesis focuses on methods for performing concurrent map­ping and localization using a feature-based approach. The concurrent mapping and localization problem is cast as a stochastic estimation problem. Based on Kalman filtering techniques, augmented stochastic mapping is introduced as a method for performing concurrent mapping and localization in realistic scenario simulations and experiments. The role of data association ambiguity, track initiation and track dele­tion in the presence of uncertainty and non-linear system dynamics are addressed. A novel approach is introduced to overcome the computational complexity inherent in mapping large areas with many features. Adaptive concurrent mapping and local­ization based on choosing the robot's action so as to maximize the expected Fisher information is introduced in order to achieve improved performance. Results from simulations, land and underwater experiments, and post-processing of oceanic data are presented to demonstrate the validity of the proposed approaches. Once a region is mapped and localization information is available, planning collision free trajecto­ries from the current position to the goal position is important for reliable mobile robot operations. In this context, a novel path-planning algorithm based on har­monic potentials is introduced for performing path-planning and obstacle avoidance in dynamic environments.