Analysis of Large-Scale Real-World Driving Data for Prediction of Vehicle Dynamics on Arbitrary Driving Routes

Masanori Okamoto (Toyota Systems)･Hideaki Bunazawa･Tomohiro Yoshimura (Toyota Motor)･Thomas Bernes Lasserre (Toyota Motor Europe)･Shigeki Hasegawa (Toyota Motor)

Using large-scale real-world driving data collected from connected vehicles, the effects of vehicle type, region, and road conditions on vehicle dynamic characteristics (e.g., vehicle speed and acceleration) were analyzed. The obtained characteristics are expected to enable highly accurate prediction of driving behavior along arbitrary driving routes and to be applied to simulations of power performance, fuel economy, and durability.

Machine-Learning-Based Quantification of Subjective Evaluation for Vehicle Cornering Behavior

Ryoya Kanahori･Tomohiro Shimizu･Satoru Sugiyama･Makoto Okamoto･Takafumi Asano･Takashi Kaneko (SUBARU)

The assessment of vehicle dynamic performance has conventionally relied upon subjective evaluation by expert drivers, and translating such qualitative judgments into quantitative inputs for vehicle design and control has required substantial engineering effort. In this study, we developed an automated analysis framework that combines a Random Forest model with similarity assessment in the feature space, using measured driving data and subjective rating scores. The proposed approach enables rapid and reproducible derivation of performance indices that correspond to highly rated vehicle behavior, thereby establishing a foundational technology for improving the efficiency of vehicle dynamics design.

AI Image Generation for Safety Evaluation of Automated Driving Systems

Nobutoshi Ozaki･Tetsuya Niikuni (NALTEC)

Evaluating the safety of automated driving systems requires verifying whether such systems accurately recognize various situations in traffic. Datasets consisting of only real-world images require extensive driving data collection to capture rare situations, such as edge cases. In this study, we focus on object detection AI for traffic images and investigate a method for generating synthetic test images using generative AI.

Automation of test scheduling using mathematical optimization

Toshihiro Watanabe･Hiroshi Yoshimoto･Toshiki Terabe･Yuya Osamura･Junichi Abe (Toyota Motor)･Kosuke Suzuki･Tetsuro Sakamoto･Tomoko Ikuta (Fixstars)

To address the reliance on the experience and know-how of seasoned veterans for test planning, we introduced mathematical optimization to enable anyone to quickly create optimal test plans. By identifying the necessary equipment and personnel information for testing and formulating the tacit knowledge and rules of seasoned veterans into mathematical equations, we were able to not only create plans quickly but also analyze bottlenecks in test execution.