Maximum-Leaf Spanning Trees for Efficient Multi-Robot Recovery with Connectivity Guarantees.

This paper presents a self-stabilizing distributed algorithm for the recovery of a large population of robots in a complex environment-that is, to gather them all in a goal location. We assume the robots do not have a map of the environment, but instead use short-range network communications and local sensing to physically route themselves towards the goal location. Since the robot's motion can disrupt robots network communication and localization in complex environments, we desire an algorithm that can maintain connectivity while preserving efficient operation. Our approach constructs a spanning tree for physical routing, but only allows the leaves of this tree to navigate the goal location. This distributed maximum-leaf spanning tree (DMLST) ensures connectivity, while providing an efficient recovery by allowing the maximum number of robots to be mobile. We present empirical results on the competitive ratio of the DMLST and it is very good, approaching the optimal solution for our communication network. DMLST recovery has been tested in simulation, and implemented on a system of thirteen robots. While a basic recovery fails in all experiments, the DMLST recovery succeeds efficiently in most trials.