CALU is a multi-robot collision avoidance system based on the velocity obstacle paradigm. In contrast to previous approaches, we alleviate the strong requirement for perfect sensing (i.e. global positioning) using Adaptive Monte-Carlo Localization on a per-agent level.