Evaluation of 2D Localization Performance in Wheeled Robots Based on the Fusion of Odometry and Absolute Positioning Systems

摘要

Accurate localization is essential for wheeled robots in structured environments. This study evaluates the performance of an EKF-based sensor fusion approach that combines odometry with an Absolute Positioning System (APS) under different configurations, analyzing how APS setup variations affect localization accuracy. A line-following, differentially driven wheeled robot was used, providing odometry data. The Ground Truth (GT) database consists of predefined, precisely measured trajectories on which the robot was deployed. Based on these trajectories and odometry readings, APS data were artificially generated, providing virtual measurements with different noise levels. Five trajectories were analyzed, including multiple straight-line segments with directional changes and one semi-circular path. The APS was tested at two update frequencies (1 Hz and 5 Hz) and five noise levels (20,40,60,80, and 100 mm standard deviation (STD)). Performance was evaluated using Root Mean Square Error (RMSE), STD, and Maximum Absolute Error (MaxAE). Results show that EKF-based fusion improves accuracy on straight trajectories with sudden direction changes, while for the semicircular path, odometry alone performs comparably. Higher APS update frequencies enhance localization, while increased noise degrades it. To achieve RMSE lower than 50 mm, the noise level must not exceed 20 mm for 1 Hz APS and 40 mm for 5 Hz APS.