Detection Of Localization Failure Using Logistic Regression

Monte Carlo localization (MCL) is a sample-based approach for representing probability density for the pose of a robot. MCL is widely used for mobile robots because of its robustness with respect to sensor noise. On the other hand, MCL can fail to estimate a pose of a robot if objects block measuring range of a laser sensor of the robot. Even though MCL fails to estimate the pose of the robot, MCL does not stop to estimate the pose because MCL does not have a self-diagnostic function. Therefore, detection of localization failure is essential for a mobile robot application. In the present paper, we propose a novel approach that detects the localization failure of MCL through logistic regression. The proposed approach has two advantages. First, it can detect whether position errors is larger than 0.15 m with a high accuracy. Second, the proposed method can detect localization failure by means of a statistically modeled equation. Moreover, as an example of an application using the probability of localization failure, we have proposed a hybrid localization scheme with MCL and laser odometry. The practical effectiveness of the proposed scheme is verified through experiments.