Indoor positioning and navigation have become a prerequisite for many applications and services. Currently available solutions are mainly based on radio beacon networks deployed in the infrastructure or a-priori mapping of signal characteristics. The first approach is costly and requires numerous calibrations, and the second assumes that the signals are stable over time, which is rarely true. Given these limitations, the use of inertial signals offers the promise of a solution that requires no infrastructure deployment, no knowledge of the map, is functional without mobile cellular networks coverage and therefore accessible to a larger population. It has become the backbone of pedestrian localization systems but must be fused with other measures to compensate for its measurement errors. The inertial sensors embedded in wearable devices are first of low quality and they require continuous calibration “tricks”. Second, they measure many human motions, which are not those we are trying to estimate to track the human gait. In this context, what accuracy is achieved by the recent inertial signals processing methods? Can we measure complex human movements even when the wearable device is worn by the pedestrian on different parts of the body? Is the hybridization of inertial signals with other data still needed to ensure high accuracy? What is the best strategy to label time series of inertial data for learning human gait features with artificial intelligence: physics or human interpretation? I will talk about recent developments in indoor pedestrian localization with inertial measurements, including artificial intelligence approaches, whether for professional applications where a 6.8m after 2’500m 3D accuracy on 3 indoor levels (ANR/DGA MALIN competition) was recently achieved or for the general public with wearables where learning issues to gain robustness in everyday life will be presented.”