Evaluating failure in terrain coverage by autonomous agents

Abstract

The work on terrain coverage algorithms has been shown to be applicable to many real-world problems. Examples range from battlefield tactics to planetary exploration and from spell-checking to vacuuming. The general idea behind terrain coverage lies on exploring a defined area/environment as much as possible. Previous works have demonstrated the effectiveness of algorithms in various scenarios such as unobstructed and obstructed terrains, and terrains that contain special, hard-to-find, locations such as rooms accessible only by a small entrance. However, these evaluations fall a little short because they do not consider the fact that agents covering an area are prone to failures. For example, a robot in a planetary exploration task can break down and be permanently or intermittently unavailable. This paper attempts to close this gap by evaluating terrain coverage algorithms under three main failure models. The failure models themselves are defined here based on an empirical analysis of how agents fail in real-world applications.