Experimental Validation of NDT-AMCL: a Precise and Reliable Localizer for Mobile Robots in Human Crowds Using Normal Distribution Transforms

The number of applications for autonomous mobile robots is steadily increasing, with a growing trend toward deploying these robots in close proximity to humans and in environments with a high density of people. In particular, indoor settings present a unique challenge for localization algorithms, as most rely on sensors that orient the robot based on static features of the surrounding environment. The presence of dynamically moving people disrupts this process, making accurate localization more difficult. Widely used algorithms, such as Simultaneous Localization and Mapping (SLAM) and Monte Carlo Localization (MCL), encounter limitations when the robot is surrounded by large numbers of people or operating within a dense crowd. As a result, the development of more robust and accurate localization algorithms that can reliably function in human-crowded environments is essential. In this paper we show the applicability of our new algorithm NDT-AMCL in the real world. NDT-AMCL is an extension of the popular Adaptive Monte Carlo Localization (AMCL) algorithm that integrates Normal Distribution Transforms (NDT) to improve performance. Through extensive realworld experiments, we evaluate the effectiveness and applicability of NDT-AMCL in crowded environments. Our findings show that NDT-AMCL significantly outperforms standard AMCL in human-free settings and, more importantly, demonstrates superior performance in densely populated environments compared to AMCL's performance in human-free scenarios. In addition to its improved performance, NDT-AMCL is just as easy to use as AMCL, enabling simple integration into robotics projects. To foster further research and development, we have made NDT-AMCL available to the research community as a straightforward extension of the AMCL package within ROS2.