Recognition of traffic information is essential for high-level urban autonomous driving. Especially, the traffic light is the most important signal because it controls pedestrian and traffic flow. There are some methods such as V2X, image processing, and deep-learning to recognize traffic lights. This paper presents deep-learning based detection and tracking algorithm for traffic light signal recognition. We use the deep learning model YOLOv3 to classify traffic light signals in one stage for fast and accurate detection. Moreover, to make a robust deep learning model, we gather plenty of data from open-source data sets and vehicle front camera videos. Then, we combine the object tracking method using IOU to give reliability to continually detected signals to ignore sudden detection failures or misdetection. The experiment result from the approaching intersection scenario demonstrates the decent performance of traffic light recognition algorithm.

With recent rise in the deployment of vehicles equipped with Advanced Driver-Assistance Systems (ADAS), there has been a lot of researches on the effects of the Autonomous Emergency Braking (AEB) system for passenger injury. From this perspective, the study developed scenarios for AEB activation and simulated passenger injuries. The research focused on the analysis of how three primary variables-the seating angle of the passenger, the speed of the vehicle, and the timing of airbag deployment-affect the severity of passenger injuries. The analysis of injury data and the application of machine learning models are conducted to predict the optimal timing of airbag deployment based on specific seating angles of passengers and vehicle speeds. The objective is to explore the feasibility of adjusting the timing of airbag deployment in real-time within AEB systems. Based on machine learning, the prediction model can improve the passenger injury accroding to passenger seating angle and vehicle speed. This study shows the potential for the adjustments in airbag deployment timing to improve the effectiveness of AEB systems to mitigate injuries during accidents.

This study describes a system parameter-free adaptive path tracking control algorithm for autonomous mobility based on input-coupled error dynamic model. Various model-based control algorithms for path tracking of autonomous mobility require relatively accurate mathematical model and system parameters. Due to the variability of mobility system parameters caused by environmental or system condition changes, accurate parameter derivation or real-time estimation algorithms are necessary to ensure effective control performance. In order to overcome the mentioned limitation, coupled dynamic model that consists of control errors and inputs has been designed and real-time coefficient estimation algorithm has been designed using recursive least squares with multiple forgetting. The coefficients in the coupled dynamic model is designed to be estimated by using control errors and inputs. The estimated coefficients are used to derive steering control inputs. The cost function has been designed using control errors and the Lyapunov direct method has been used to derive front and rear steering control inputs for path tracking. The proposed path tracking the proposed control algorithm can be used as a universal path tracking algorithm that does not need system parameters for various autonomous mobility platforms.

In forklift-related accidents, two representative accident types, jammed and impact, were often caused by obstructed view. Around-view monitoring systems (AVM), commonly used in passenger cars was expected to resolve the obstructed view problem in forklifts. This study aimed to assess their potential benefits in forklift safety. By driving a forklift with the around-view monitoring system (AVM) and the forklift without the AVM, the braking times were measured and compared. The controlled test was conducted with eight drivers, each performing 32 trials (16 with AVM, 16 without AVM) involving forward and backward driving. Braking times were measured in response to one of eight lights switched on, which were positioned along the predetermined path. As a result, no significant difference was observed in average braking times depending on the AVM condition. However, outlier analysis revealed twice as many outliers in test condition without AVM, compared to the condtion with AVM. Therefore, around-view monitoring systems in forklifts could help reducing the number of unrecognitions, thereby preventing forklift-related accidents from obstructed view.

This paper presents a steering control system model for synchronized driving of two vehicles to increase the freedom of cargo transportation. The two autonomous vehicles drive side by side in the lateral direction to safely transport one large cargo. The overall steering control algorithm for synchronization consists of two parts. The first part preemptively determines the vehicle speed and steering angle for the two vehicles to have the same instantaneous rotation center using the Ackerman steering mechanism. However, the gap between the vehicles is not maintained because it is impossible to accurately achieve the kinematically required vehicle speed and steering angle of each wheel at every moment. To address this issue, the two vehicles form a master-slave relationship, allowing the slave vehicle to track the master vehicle. The second part of the steering control algorithm generates a trajectory of the virtual vehicle having a constant lateral offset from the master vehicle and determines an additional corrective steering angle of the slave vehicle to reduce the position error with the virtual vehicle. The position error dynamics can be described by a bilinear control system. The error state feedback steering control system was designed and the control performance was verified via computer simulation studies.

This study aims to investigate the influence of robotaxi service experiences on perceived safety in autonomous driving systems. Robotaxis, as a key application of autonomous driving technology, hold significant potential to revolutionize urban mobility and transportation systems by enhancing convenience, reducing traffic congestion, and mitigating environmental impacts. The analysis is based on a pilot robotaxi service conducted in Gangneung, South Korea, where Level 4 autonomous vehicles operated along a 24-kilometer fixed route connecting major tourist destinations. Changes in perceived safety before and after the service experience were evaluated using a Repeated Measures ANOVA, which indicated a statistically significant improvement in perceived safety. This underscores the pivotal role of real-world experience in enhancing user trust and acceptance of autonomous vehicle technologies. The survey further identified potential benefits, including reductions in air pollution and improvements in public transportation accessibility and mobility efficiency. The findings suggest that user-centric pilot programs and targeted interventions are essential for fostering trust and overcoming social acceptance barriers. This study provides critical insights for advancing the deployment of autonomous driving systems and their integration into urban mobility infrastructures, emphasizing the importance of addressing user trust and perceived safety.

Recently, the number of electric vehicles has increased due to carbon reduction policies. As the number of electric vehicles registrations increases, research is being conducted to evaluate the batteries installed in electric vehicles. In this study, battery evaluation was performed using real-time driving data of electric vehicles. The RLS (Recursive Least Squares) algorithm and the DEKF (Double Extended Kalman Filter) algorithm were applied using MATLABTM, and BMS data and simulation data were compared for the battery SOC and SOH. As a result, the SOC error rate was confirmed to be less than 2.5% and SOH less than 5.0%. We plan to continuously improve the accuracy of the algorithm by collecting operating data for more than one year.