A Kalman-Particle Hybrid Filter For Improved Localization of AGV In Indoor Environment

Automated Guided Vehicles (AGV), are becoming a popular choice for a wide variety of industries ranging from large-scale warehouses and automotive assembly plants to the smaller healthcare industries. Due to this versatility of operational areas, their ability to self-locate becomes very critical. However, the traditional localization approach such as classic Monte Carlo Localization (MCL) fails as it relies on the noisy measurements from encoders. An alternative approach called Iterative Closest Point (ICP) uses LIDAR to avoid unbounded noises measurements but fails to generate consistent samples in symmetrical environments. In this paper, an Extended Kalman Filter (EKF) based proposal distribution is introduced that combines encoder measurements with the LIDAR data to overcome the limitations in the classic MCL. The EKF generates samples efficiently by incorporating an adaptive observation covariance estimation solution. The proposed method is implemented in Robotic Operating System (ROS) and the tests are performed in a simulated environment, generated by the Gazebo simulator. The results of the proposed method show an overall improvement in the localization accuracy in comparison to the classic localization frameworks.