Abstract
In this paper, we propose failure detection and recovery method for localization by using Particle Filters (PFs). PF finds a correct position from limited number of samples. Therefore, PF cannot detect that there is not a correct position because all possible positions are not evaluated. We consider the problem that is arose by the relative evaluation. Our proposal is embedding absolute evaluation to PF by utilization of ratio between estimated maximum likelihood and l ikelihood of samples. When the ratio is low value, proposed method detects the event as the failure. Experimental results show proposed method can robustly detect failures and recover localization by using GPS.
