Ultra Wide Band (UWB) is an emerging technology in the field of indoor localization, mainly due to its high performances in indoor scenarios and relatively easy deployment. However, in complex indoor environments, its positioning accuracy may drastically decrease due to biases introduced when emitters and receivers operate in Non Line-of-Sight (NLOS) conditions. This undesired phenomenon can be attenuated by creating, a priori, a map of the measurement error in the environment, that can be exploited at a later stage by a localization algorithm. In this paper, the error map is the result of a calibration process, which consists of collecting several measurements of the localization system at different locations in the environment. This work proposes the leveraging of mobile robots in order to automatize the calibration process with the ultimate purpose of improving UWB-based people localization in a realistic indoor environment. The whole process exploits existing algorithms in the field of robot localization conveniently adapted in order to address our use case and technology. Experiments in real environments of incrementally increasing complexity show how the average localization accuracy can be improved up to 50% by adopting this method.