Investigation on the Effect of Low-Dynamic Environments in NDT-based SLAM

Self-localization is a crucial capability for a robot in all its applications, nonetheless autonomous vehicles. They must be able to properly locate themselves inside dynamic environments. SLAM systems of autonomous vehicles usually rely on laser sensors, or lidars, to sense its surroundings, for the only purpose of self-localization. These systems perform poorly in low-dynamic environments, which contain objects whose dynamics is much slower than the robot's.
Understanding the performances of lidar-based localization systems in low-dynamic environments has been the main focus of this work. To do so, an experimental methodology, in which a robotic platform equipped with a lidar was manoeuvred in an artificial dynamic environment, was performed. The collected laser scans were then processed by a lidar-based localization system. In doing so, the effect of the low-dynamic entities was assessed. Results showed that low-dynamic objects carry vital information for the robot's self-localization, as long as their detection through lidars leads to sets of measurements that partially resemble the references the system had previously acquired about the navigated environment.
These results were finally used to propose an improvement of the used system, that makes uses of a semantic segmentation of the collected laser measurements.