Various machine learning models have been proposed to predict injuries to vehicle occupants from crash conditions, and optimized and evaluated for model accuracy using binary classification performance metrics. However, performance metrics for injury probability prediction have not been utilized to develop injury prediction models. Therefore, this paper developed injury probability prediction models using evaluation metrics to evaluate the probability prediction performance of injury prediction models and to verify the validity of injury probability predictions. The model constructed using a random forest performed better than the conventional injury prediction model constructed using logistic regression.

View full abstract

In an effort to improve automated driving and advanced driver assistance systems (AD/ADAS) performance, the industry is gradually moving toward greater vehicle autonomy. In this context, particular emphasis is being placed on the use of in-vehicle sensors and cameras. While these systems are primarily intended to replace or support the driver with invehicle devices, the ability to actively perceive non-visible areas can significantly improve vehicle safety and optimize route decisions. Traditional navigation maps (SD-MAP) have served their purpose, but often lack the necessary dynamism and accuracy; Vehicle-to-Everything (V2X) technology, particularly in the area of Vehicle-to-Network (V2N), is a growing area of interest due to the proliferation of smart phones and offer promising alternatives that leverage the strength of cellular networks. This study will evaluate the utility of smartphones as a central element of vehicle sensing and investigate their potential to provide critical lane-level recognition data for next-generation AD/ADAS applications. Through a comprehensive system architecture, smartphone GNSS data is processed to estimate lane-specific traffic conditions. In addition, current limitations and potential enhancements with respect to the use of OpenStreetMap (OSM) data will be detailed. While the accuracy of lane identification based solely on data from smartphones remains a challenge, this paper explores possible countermeasures, highlighting the integration of various data sources to achieve a holistic vehicle control system.

View full abstract

This study examined the feasibility of improving traffic flow on urban highways using AD/ADAS and connected systems. The focus was on congested merging areas with the aim of maintaining the speed immediately after merging. The effectiveness of lane-based vehicle relocation and speed control measures was evaluated to achieve this goal. This study also considered realistic specifications for connected systems, considering constraints such as cost limitations. The feasibility of improving traffic flows through strategies such as lane utilization management and speed control was investigated and potential new challenges were identified.

View full abstract

In recent years, autonomous driving has been attracting great attention. However, autonomous driving in today's technological level still has problems in operating in complex environments such as urban areas, disaster sites, and scenes of accidents. In such situations, teleoperation technology is expected to serve as a backup for autonomous driving systems. A major technical challenge in remote driving is steering instability caused by communication delay and poor visual information. In this paper, we propose a novel predictive display that can improve the steering stability of remote driving. Experimental results with five subjects showed notable improvement of driving performance.

View full abstract

A theoretical virtual machining method is proposed to identify the optimal machining center-line. The process initiates with a meticulous scanning and reverse reconstruction of forged crankshafts model, which is then discretized into numerous pieces for assessing imbalance force and moment. An iterative process is introduced to search for the mass center-line of the crankshaft, followed by a virtual machining method to obtain the virtual final product. An optimization procedure is employed to accurately identify the distances between the optimal machining center-line and the mass center-line. It can be theoretically applied in CNC crankshaft mass centering machines to determine offset values.

View full abstract