This paper investigates an infra-structure free global localization of a group of communicating mobile agents (e.g., first responders or exploring robots) via an ultra-wideband (UWB) inter-agent ranging aided dead-reckoning. We propose a loosely coupled cooperative localization algorithm that acts as an augmentation atop the local dead-reckoning system of each mobile agent. This augmentation becomes active only when an agent wants to process a relative measurement it has taken. The main contribution of this paper is addressing the challenges in the proper processing of the UWB range measurements in the framework of a loosely coupled cooperative localization. Even though UWB offers a decimeter level accuracy in line-of-sight (LoS) ranging, its accuracy degrades significantly in non-line-of-sight (NLoS) due to the significant unknown positive bias in the measurements. Thus, the measurement models for the UWB LoS and NLoS ranging conditions are different, and proper processing of NLoS measurements requires a bias compensation measure. We also show that, in practice, the measurement modal discriminators determine the type of UWB range measurements should be probabilistic. To take into account the probabilistic nature of the NLoS identifiers when processing UWB inter-agent ranging feedback, we employ an interacting multiple model (IMM) estimator in our localization filter. We also propose a bias compensation method for NLoS UWB measurements. The effectiveness of our cooperative localization is demonstrated via an experiment for a group of pedestrians who use UWB relative range measurements among themselves to improve their shoe-mounted INS geolocation.