Failure detection for laser-based SLAM in urban and peri-urban environments

Abstract

Simultaneous Localization And Mapping (SLAM) is considered as one of the key solutions for making mobile robots truly autonomous. Based mainly on perceptive information, the SLAM concept is assumed to solve localization and provide a map of the surrounding environment simultaneously. In this paper, we study SLAM limitations and we propose an approach to detect a priori potential failure scenarios for 2D laser-based SLAM methods. Our approach makes use of raw sensor data, which makes it independent of the underlying SLAM implementation, to extract a relevant descriptors vector. This descriptors vector is then used together with a decision-making algorithm to detect failure scenarios. Our approach is evaluated using different decision algorithms through three realistic experiments.