Swarming Out of the Lab: Comparing Relative Localization Methods for Collective Behavior

Efficient collaboration within a robot swarm hinges on the precise localization of swarm members relative to their neighbors. However, in real-world scenarios, such as indoor GPS-denied environments, access to accurate global localization systems is typically limited, and relative localization poses challenges due to the absence of a global reference frame. This paper compares the localization accuracy of three methods: IR-based, visual-inertial, and ultra-wideband localization systems. We evaluate these systems to ascertain the relative localization accuracy of neighboring robots engaged in collective behaviors. We develop a simulation model for the three localization systems and conduct accuracy studies. Furthermore, we deploy two swarms, one consisting of five flying robots and one consisting of five ground robots performing three distinct behaviors to validate the simulation experiments. Through simulation and robot experiments, we present the characteristics of each system, including estimation accuracy, deployment cost, communication overhead, and behavior performance accuracy.