Navigation under Rank-Deficient Localization for Nonholonomic Ground Vehicle
-A Special Spiral Curve Method-

Wei Wang･Kimihiko Nakano (The University of Tokyo)

This study proposes a vehicle navigation method using special spiral curves for rank-deficient localization, where only lateral position feedback is available. The proposed approach enables accurate navigation without odometer-based positioning and avoids curvature breakpoints. Experimental results demonstrate the feasibility of navigation under sensor-limited conditions and show that the method is potential to reduce path deviation.

Design and Development of a Photorealistic Simulation Platform for Model-Based Development of End-to-End Autonomous Driving

Keigo Nakamura･Miyu Yokoyama (Aisin)

CG-based autonomous driving simulation faces challenges such as domain gaps with real-world imagery and high modeling costs. To address these issues, this study proposes a simulation platform that jointly simulates 3D Gaussian Splatting and vehicle dynamics. By tightly integrating photorealistic scene representation with physically consistent vehicle motion, the proposed approach achieves both visual realism and motion accuracy. The platform is evaluated through experiments, confirming its effectiveness for model-based development (MBD) of autonomous driving systems.

Evaluating Occluded Pedestrian Detection and Enhancing Distant Recognition via Virtual Telephoto Cameras in Autonomous Driving

Yitong Wang･Jishu Miao･Yukiya Hattori･Tsubasa Hirakawa･Takayoshi Yamashita･Hironobu Fujiyoshi (Chubu University)

To enhance the safety of autonomous driving and improve the detection accuracy for distant objects, we propose a Multi-View BEVFormer that combines conventional camera views with a view focused specifically on distant regions. By evaluating the effectiveness of this proposed method, we demonstrated that it achieves highly accurate detection of objects 100 meters away, which are traditionally difficult to detect using conventional approaches.

Application of High-Precision LiDAR Recognition and Automated Inspection for the Introduction of Infrastructure-based Autonomous Driving Systems into Vehicle Inspection Factory

Shinya Hozumi･Takeshi Kanou･Yuki Okamoto･Yuhei Nagafuchi･Kento Iwahori･Noritsugu Iwazaki･Takuro Sawano･Yuta Kishioka (Toyota Motor)･Seiya Takata･Kenta Azuma (DENSO)

To address the challenges of labor shortages and to enhance productivity in vehicle manufacturing plants, we are developing an infrastructure-based autonomous driving system. In environments where humans and equipment coexist, and under conditions with external disturbances, we have developed self-localization and object detection functions based on high-precision LiDAR, replacing conventional camera-based approaches. Furthermore, we are investigating the feasibility of automated inspection enabled by more precise vehicle control.

Active Misalignment Correction System for Autonomous Driving Radar Using IMU/MAG Sensors and Actuators

Sukbom Son (Hyundai Motor)･Juhwan Park･Wanho Son (Ecoplastic)･Jin Young Yoon･Dongha Kim･Hongheui Lee･Sungho Park･Soobok Kim (Hyundai Motor)

Automotive radar sensors are a critical component of Advanced Driver Assistance Systems (ADAS), and their misalignment directly compromises autonomous driving safety. When integrated into the bumper rather than the front-end module (FEM), angular deviations are difficult to manually correct due to restricted physical access, and the exposed position makes the sensor inherently susceptible to impact-induced misalignment. In this study, we propose an automatic self-leveling system that detects radar misalignment via IMU and magnetometer (MAG) sensors and physically corrects the deviation using a custom-designed 2-DOF actuator, eliminating the need for manual calibration.

Tightly Coupled LiDAR-IMU-Odometry Fusion with Continuous-Time Trajectory Representation for Robust Ego-Vehicle Localization

Naoki Akai･Yasuhiro Akagi･Takayuki Morikawa (Nagoya University)

LIO can be used with only a LiDAR and an IMU, while odometry, which represents vehicle motion characteristics, is effective for improving robustness in vehicle applications. This presentation describes a robust vehicle localization technique that extends conventional discrete-time state estimation systems to continuous-time state estimation systems, enabling seamless fusion of LiDAR, IMU, and odometry according to their respective characteristics.