A comparative study on robust localization: Fault tolerance and robustness test on probabilistic filters for range-based positioning

As autonomous robots are becoming more and more involved in service and domestic scenarios, the deep interaction with human actors imposes harder constraints on safety and response to adverse situations. We propose fault tolerance and robustness tests for indoor beacon-based localization. Applying concepts of robust statistics, we provide a set of fault hypotheses and environmental perturbations, in order to study the performance of localization systems despite faults or external attacks. Under this set we compare five probabilistic techniques: extended Kalman filter localization, Kalman filter localization with likelihood test, Monte Carlo localization, Augmented Monte Carlo localization and our model, called Probabilistic Shaping. Performance evaluation is managed in terms of transient and steady state errors analysis, simulating filters estimation in a realistic scenario. We derived some important drawbacks on filters behavior when systematic errors occur. The results are particularly significant when dealing with low-cost radio frequency beacons, in which inferior measurements quality becomes part of normal working conditions and is no longer a rare contingency.