Development of a Map-based Detection Method of Road Markings Degradation by LiDAR-camera Fusion

Akisue Kuramoto (Institute of Science Tokyo)･Ryo Yanase･Keisuke Yoneda･Naoki Suganuma (Kanazawa University)

Proper recognition of road markings' degradation state is important for road maintenance management and the reliability of assistance or autonomous driving systems. This paper proposes a map-based detection method of road marking degradation using quantitative indicators, such as the reflectance and visibility of ​​road markings, obtained by fusing a camera and LiDAR.

Collision Risk-map Robust to Errors in Obstacle's Predicted Trajectory for Trajectory Planning of Autonomous Vehicle

Yuki Kasai･Kazuhiro Sorimachi･Nobuhiro Watanabe･Wataru Yoshiuchi･Nobuhiro Hayashida (Isuzu Advanced Engineering Center)

Trajectory planning based on a risk-map expressing risks caused by obstacles is suitable for urban autonomous driving. This is because the map treats diverse elements in a uniform manner. In a previous work, a risk-mapping method of evaluating spatio-temporal proximity was proposed to express collision risks with moving obstacles. However, this method is susceptible to deviation of the obstacle’s predicted trajectory when applied to actual vehicles. In this study, we apply a binary Bayes filter to the method. Through actual vehicle experiments, we confirmed that our method improves noise robustness and enables trajectory planning that avoids collisions with moving obstacles.

Alignment of Map Coordinate Systems Toward Fusion of Multiple Localization Technologies

Ryoma Kakimi･Taishi Shiotsuki (Toyota Motor)

Fusion of multiple localization technologies and fleet management by control systems are often performed in the areas of autonomous mobility. These technologies often have their own coordinate system, and accurate alignment of these systems is necessary for stable system integration. This paper reports on a technique that uses reference anchor points of each coordinate system to statistically derive the coordinate transformation to enable accurate alignment.

Data-driven Planner for Autonomous Driving Systems Considering Application to Vehicles with Various Specifications

Akira Ito (Aichi Institute of Technology)･Ken Kinjo･Kenta Mukoya･Yuki Asada (DENSO)

A data-driven planner for autonomous driving systems does not work properly for vehicles with different specifications from those used during training, and thus requires re-training for each vehicle.
In this study, we propose a method that combines a data-driven planner and model following control to enable the planner to be applied to vehicles with various specifications without re-training.