A survey on digital twin-assisted intelligent vehicle localization

초록

Localization modules play an important role in ensuring the secure operation of intelligent vehicles (IV) and accelerating the development of driving technologies. To safely drive an intelligent vehicle, an exact data registration is necessary. Map drift, scan-to-map alignment, high computational load, sensor noise, and calibration are the most prominent issues that cause poor localization performance for intelligent vehicles. Therefore, by presenting a digital twin-assisted data registration (DT-ADR) technique, this research attempts to mitigate that gap. The purpose of this study is to show how to implement the DT-ADR technique to help improve the existing light detection and ranging (LiDAR) data registration technique for more accurate vehicle localization and driving capabilities. We first investigate traditional data registration techniques that address the challenges for scan matching localization. We then present the proposed DT-ADR technique and discuss a pose selection technique. Next, we demonstrate an implementation of the proposed DT-ADR technique using AWSIM cosimulation, Autoware universe, and ROS2 virtual environments. To verify the effectiveness of the description of the localization approach, a case study is also conducted to analyze the pose estimation of IV. Lastly, an initial result is evaluated for the proposed DT-ADR technique, which reduces the orientation root mean squared error by an average compared to the existing LiDAR data registration technique based on normal distribution transforms. This work could be utilized as a testbed for future research that attempts to include advanced localization features for IV driving.

키워드