Active robot localization with macro actions

Abstract

Abstract-We propose a novel method for solving the localization problem, particularly useful in self-similar environments. Localization is the task of determining a robot's position with little or no prior knowledge about its initial location. We introduce a computationally efficient active strategy for the localization task. The proposed method also automatically generates optimal macro actions giving the robot the ability to localize in all environments. Combining the active strategy and macro actions, the method introduces an efficient strategy for determining a robot's position in any environment where localization is possible.

Active Domain-Invariant Self-Localization Using Ego-Centric and World-Centric Maps

“…Active VPR plays a particularly important role in scenarios in which robots navigate featureless environments. Typical examples of such scenarios include indoor VPR tasks in featureless passageways and classrooms, and outdoor VPR tasks in featureless road environments [5,6,7]. Other important scenarios include active SLAM tasks, in which an alternative VPR task known as active loop closing is considered to recognize revisits using an incomplete environmental map that is being built and often featureless [18,19,20].…”

“…However, such passive formulation is ill-posed and its VPR performance is significantly dependent on the presence or absence of landmark-like objects in the view image [4]. To address this ill-posedness, several studies have extended the passive VPR task to an active multiview VPR with NBV planning [5,6,7]. Active VPR is formulated as a state-to-action mapping problem that aims to maximize the VPR performance while suppressing the expected costs of action and sensing at future viewpoints.…”