Extended Particle Filter Based on the Maximum Correntropy Criterion for Mobile Robot Localization in Indoor Environments.
Particle Filter, Maximum Correntropy Criterion, Mobile Robot Localization, Non-Gaussian Noise, State Estimation.
The increasing automation in indoor environments, such as Industry 4.0, logistics, and assistance services, spurs the demand for Autonomous Mobile Robots (AMRs) with robust and accurate localization capabilities. Autonomous navigation in these scenarios is challenging due to the absence of Global Navigation Satellite System (GNSS) signals and the presence of non-Gaussian noise and outliers in sensor data. Probabilistic approaches, such as the Particle Filter (PF), are a class of solutions used to estimate the robot's position in the environment and in algorithms like SLAM, but their effectiveness is compromised by this noise, which can lead to particle degeneracy, sample impoverishment, and consequent filter divergence.
This work proposes the optimization and adaptation of a Maximum Correntropy Criterion-based Extended Particle Filter (MCEEPF) to achieve robust mobile robot localization. The strategy consists of enhancing the importance density function, which guides the particle sampling process. To this end, an Extended Kalman Filter, based on the correntropy criterion, is employed to generate the distribution's parameters. This approach aims to suppress the impact of outliers and steer new samples towards high-likelihood regions in the robot's state space. The objective is, therefore, to refine and apply this methodology to the specific domain of indoor mobile robotics, validating the gains in accuracy and robustness over the conventional Particle Filter approach, especially in scenarios with non-ideal noise.