Simulation environments for autonomous mobile robots often include simplified sensors and associated environment models. The output of these sensors are essential for the evaluation of robotic capabilities. Invalid assumptions in sensor models and the associated environment models thus lead to false assumptions of the subsequent algorithms, whose decisions the robot is later defeated by. In order to bridge this gap, we propose a LiDAR sensor model, which uses a physics based approach: Modifications to the XYZ-position and intensity values are conducted with the goal in mind to simulate real measurements as close as possible. Additionally, this model also takes drop-outs into account to further increase the realism of the simulation model. This behavior is also modeled by analyzing the reflected power of the incident LiDAR signal, thus conforming to fundamental principles of physics. The sensor model that has been introduced can be computed in real-time using only the CPU. The impact of the adapted LiDAR model is evaluated by an application scenario, namely the localization task using Google's Cartographer.
What can we Learn from Virtual Sensor Models for Self Localization and Mapping for Autonomous Mobile Systems?
2023-09-24
5177556 byte
Conference paper
Electronic Resource
English
Autonomous Navigation—Self-localization and Mapping (SLAM)
Springer Verlag | 2015
|What can we learn from autonomous level-5 motorsport?
Springer Verlag | 2018
|NLMAP: visual-based self localization and mapping for autonomous underwater vehicles
Tema Archive | 2008
|