Study on Robust Self-Localization against Seasonal Environmental Changes for Automated Vehicles

Shun Nishimura･Manabu Omae (Keio University)

This study proposes a self-localization method utilizing artificial structures to improve the reliability of self-localization accuracy for long-term in locations affected by plant-induced environmental changes. In this paper, we report on the construction of a high-precision map and a self-localization algorithm for artificial structures, as well as the results of self-localization accuracy in different seasons.

Implementing Localization using 4D Imaging Radar

Sota Izumi･Keisuke Yoneda･Ryo Yanase･Mohammad Aldibaja･Naoki Suganuma (Kanazawa University)

Accurate self-localization is important to realize safe autonomous driving. Conventionally, millimeter-wave radar has been used for self-localization to ensure robustness against changes in the surrounding environment, such as snow and heavy rain. But there is an issue of estimated accuracy due to low resolution. This study aims to improve accuracy of self-localization while maintaining high robustness against environmental changes by implementing localization module which uses 4D imaging radar with high resolution.

Robust Map Matching for Environmental Changes using CNN

Kota Jimbo･Keisuke Yoneda･Ryo Yanase･Mohammad Aldibaja･Naoki Suganuma (Kanazawa University)

Localization is important to enable safe autonomous driving. When environmental changes occur, infrared reflectance from the road surface decreases due to rainy weather and blurring of white lines. Then, the localization accuracy is reduced by using a method based on infrared reflectance by LiDAR. The objective of this research is to achieve robust matching against environmental changes by using convolutional neural network, CNN.

Multi-Agent Approach for AD/ADAS Cross-Country Virtual Validation

Reza Rezaei･Ravibhai Vaghasiya･Jacob Grandmontagne･Morteza Molaei･Frank Reifenrath (IAV)

This paper introduces a novel Multi-Agent Approach coupled with high-precision environment modeling for cross-country virtual validation of AD/ADAS systems. It focuses on accurately representing country-specific traffic signs, rules, conditions, and driving scenarios, highlighting critical situations under various operating conditions.
Through the utilization of advanced AI-based multi-agent modeling, the realism and accuracy of simulated scenarios are further enhanced. The advantages and limitations of multi-agent approach, considering diverse driver behaviors, traffic rule violations, etc. for providing a comprehensive assessment of the AD functions will be discussed based on representative simulation results from traffic scenarios in Japan and Germany.

SIL, HIL, OTA – A Study on Different Usages of Simulation and Validation

Andreas Himmler (dSPACE)･Hiroki Hanaoka･Takashi Yamada (dSPACE Japan)

In the development of perception and control algorithms for ADAS (Advanced Driver Assistance System) and AD (Autonomous Driving), it is important to verify the development results at the earliest possible stage to both shorten the development period and ensure quality. The challenge is how to use both simulation and actual products while taking into account the fidelity of simulation and the flexibility of testing. This paper studies how to distinguish the use of simulation from actual products, especially when over-the-air simulation is effective in the validation of systems using radar sensors, which are indispensable for ADAS and AD applications in recent years.