Robot Swarms Decide under Perception Errors in Best-of-N Problems

Robot swarms have been used extensively to examine best-of-N decisions; however, most studies presume that robots can reliably estimate the quality values of the various options. In an attempt to bridge the gap to reality, in this study, we assume robots with low-quality sensors take inaccurate measurements in both directions of overestimating and underestimating the quality of available options. We propose the use of three algorithms for allowing robots to identify themselves individually based on both their own measurements and the measurements of their dynamic neighborhood. Within the decision-making process, we then weigh the opinions of robots who define themselves as inaccurately lower than others. Our research compares the classification accuracy of the three algorithms and looks into the swarm’s decision accuracy when the best algorithm for classification is used.