FICE: Positional Feature Influenced Covariance Estimation for Indoor Robot Localization

Abstract

In recent years, robotics has undergone remarkable advancements, expanding its applications from hazardous industrial environments to everyday tasks. A critical aspect of indoor mobile robots is accurate localization, essential for reliable task execution in unpredictable surroundings. This thesis addresses challenges stemming from unreliable lidar measurements and the introduction of newobjects in environment, focusing on enhancing the localization capabilities of indoor mobile robots.

The proposed method introduces Positional Feature Influenced Covariance Estimation (PFICE) to enhance the Extended Kalman Filter (EKF) localization system. PFICE-Enhanced EKF dynamically adjusts measurement uncertainty based on real-time lidar scan measurements. Unlike traditional methods that assume fixed measurement uncertainties, PFICE adapts to current environmental conditions and utilizes prior knowledge from a point cloud map. This adaptive approach significantly improves localization accuracy and robustness, particularly in scenarios involving sudden movements or environmental changes. The method employs feature extraction techniques to derive structured feature vectors from raw lidar range measurements. These features are then compared with a pre-computed feature map of the environment using kernel functions, guiding covariance estimation within the EKF framework.

Evaluation against a standard method across various scenarios demonstrated superior performance of the PFICE-Enhanced EKF localization system. It achieved accurate estimations in 18 out of 28 test cases, outperforming the reference method, which succeeded in only 10 cases under similar conditions. The PFICE-Enhanced EKF effectively mitigated the impact of sudden movements and environmental changes caused by the introduction of new objects, thereby enhancing localization reliability and performance.

While promising, the study acknowledges limitations such as testing in a single environment due to the lack of access to ground truth data in different environments. Further research in diverse settings is essential to validate and generalize these findings, ensuring broader applicability of the PFICE-Enhanced EKF localization system in real-world applications.