This paper presents an vulnerable road user (VRU) classification and tracking algorithm using vision and LiDAR sensor fusion method for urban autonomous driving. The classification and tracking for vulnerable road users such as pedestrian, bicycle, and motorcycle are essential for autonomous driving in complex urban environments. In this paper, a real-time object image detection algorithm called Yolo and object tracking algorithm from LiDAR point cloud are fused in the high level. The proposed algorithm consists of four parts. First, the object bounding boxes on the pixel coordinate, which is obtained from YOLO, are transformed into the local coordinate of subject vehicle using the homography matrix. Second, a LiDAR point cloud is clustered based on Euclidean distance and the clusters are associated using GNN. In addition, the states of clusters including position, heading angle, velocity and acceleration information are estimated using geometric model free approach (GMFA) in real-time. Finally, the each LiDAR track is matched with a vision track using angle information of transformed vision track and assigned a classification id. The proposed fusion algorithm is evaluated via real vehicle test in the urban environment.

This paper presents LiDAR static obstacle map based vehicle dynamic state estimation algorithm for urban autonomous driving. In an autonomous driving, state estimation of host vehicle is important for accurate prediction of ego motion and perceived object. Therefore, in a situation in which noise exists in the control input of the vehicle, state estimation using sensor such as LiDAR and vision is required. However, it is difficult to obtain a measurement for the vehicle state because the recognition sensor of autonomous vehicle perceives including a dynamic object. The proposed algorithm consists of two parts. First, a Bayesian rule-based static obstacle map is constructed using continuous LiDAR point cloud input. Second, vehicle odometry during the time interval is calculated by matching the static obstacle map using Normal Distribution Transformation (NDT) method. And the velocity and yaw rate of vehicle are estimated based on the Extended Kalman Filter (EKF) using vehicle odometry as measurement. The proposed algorithm is implemented in the Linux Robot Operating System (ROS) environment, and is verified with data obtained from actual driving on urban roads. The test results show a more robust and accurate dynamic state estimation result when there is a bias in the chassis IMU sensor.

This paper is the result of a research on hybrid tuning technology developed to improve the actual fuel efficiency and reduce emissions of in-use light-duty diesel trucks. In this study, a hybrid powertrain was constructed by inserting an electric motor between the diesel engine and manual transmission of an internal combustion engine vehicle and installing a battery. To verify the safety, a test was conducted based on the Korean tuning regulations. In particular, since there has been no case of tuning an internal combustion engine vehicle into a hybrid vehicle in Korea, it was necessary to carry out all procedures to receive tuning approval. The approval process consists of a technical review, safety verification test, and application for tuning approval. As a result, the test vehicle was approved for tuning because both the technical review and vehicle test results were suitable. Therefore, this study confirmed the safety of diesel hybrid tuning technology, and laid the foundation for the research and development of technologies to tune into an eco-friendly vehicle as well as the activation of related industries.

This study is intended to augment the Roof strength test being evaluated by IIHS (Insurance Institute for Highway Safety). In order to find solutions for increasing Roof Crashworthiness Evaluation SWR (Strengthto-weight ratio). This study introduces that Integrated Connection Structure Between Center Pillar and Roof Center Rail is proposed as a critical solution.

This paper presents a comparative study between the parameter-optimized Pacejka model and artificial neural network model for the tire force estimation. The two different approaches are investigated and compared in this study. First, offline optimization is conducted based on Pacejka Magic Formula model to determine the proper parameter set for the minimization of tire force error between the model and test data set. Second, deep neural network model is used to fit the model to the tire test data set. The actual tire forces are measured using MTS Flat-Track test platform and the measurements are used as the reference tire data set. The focus of this study is on the applicability of machine learning technique to tire force estimation. It is shown via the regression results that the deep neural network model is more effective in describing the tire force than the parameter-optimized Pacejka model.

This paper presents a longitudinal acceleration controller that can be applied to real vehicles (nonlinear and time-varing systems) with only a simple experiment regardless of the type of vehicle and the control interface structure. The controller consists of a feedforward term for fast response, a zero-throttle acceleration compensation term, and a feedback term (P gain) to compensate for errors in the feedforward term, and another feedback term (I gain) to respond to disturbances such as slope. In order to easily apply it to real vehicles, there are only two tuning parameters, feedforward terms of throttle and brake control. And the remaining parameters can be calculated immediately when the two parameters are decided. The tuning procedure is also unified so that it can be quickly and easily applied to various vehicles. The performance of the controller was evaluated using MATLAB/Simulink and Truksim's European Ben model. In addition, the controller was successfully implemented to 3 medium-sized vehicle (HMC Solati), which is composed of different control interface characteristic. Vehicle driving performance was evaluated on the test track and on the urban roads in Siheung and Seoul.

According to the Korean Traffic Accident Analysis System (TAAS), more than 200,000 traffic accidents occur every year. Also, the statistics including auto insurance companies data show 1.3 million traffic accidents. In the case of TAAS, the types of traffic accidents are simply divided into four; frontal collision, side collision, rear collision, and rollover. However, more detailed information is needed to assess for advanced driver assist systems at intersections. For example, directional information is needed, such as whether the vehicle in the car accident way in a straight or a left turn, etc. This study intends to redefine the type of accident with the more clear driving direction and path by referring to the Negligence standards used in automobile insurance accidents. The standards largely divide five categories of car-to-car/motorcycle /pedestrian/cyclist, and highway, and the each category is classified into dozens of types by status of the traffic signal, conflict situations. In order to present more various accident types for auto insurance accidents, the standards are reclassified driving direction and path of vehicles from crash situations. In results, the car-to-car accidents are classified into 33 accident types, car-to-pedestrian accidents have 19 accident types, car-to-motorcycle accidents have 38 accident types, and car-to-cyclist accidents are derived into 26 types.

The event data recorders (EDR) have been used as a device to help understand traffic accidents. With the recent development of autonomous vehicle (AV), it has become important to prepare the new EDR for AV. Therefore, the purpose of this study is to propose the direction of EDR-AV recording. First of all, the recent EDR data elements and the data elements of AV under discussion at UNECE WP29 EDR/DSSAD (Data Storage System for Automated Driving) were analyzed. The consumer complaint database in Motor Vehicle Recall Center in Korea was analyzed in order to utilize cases of domestic traffic accidents related to advanced driver assistance systems (ADAS). Consequently, problems with existing EDR were identified through unclear accident cases related to ADAS. In the future, it was proposed to record images in which the ADAS perception systems recognize the surroundings of the accident site as an EDR-AV recording item.

This study was initiated to improve of the defect investigation method using event data recorders (EDR) and suggested a solution through the regulation and system analysis of EDR. The EDR data has been used for various purposes such as the vehicle defect investigation and the traffic accident investigation. However the EDR regulation has not been updated since the implementation in 2012. "Trigger Threshold" can be used to analyze a single accident such as the frontal crash, the side crash, and the rollover. In the case of a complex accident in which a rollover accident and a crash accident occur simultaneously, it is difficult to analyze a complex accident due to current "Trigger Threshold". This study proposed the method of separating the "Trigger Threshold" into a crash accident and a rollover accident so that accidents can be analyzed using the EDR data even when a complex accident occurs. In addition, it proposed the improvement method to quickly use the data of EDR in accident reconstruction software.

Recently, Korea has established a plan for the supply of hydrogen vehicles and is promoting the expansion of the supply. Risk factors for hydrogen vehicles are hydrogen leakage, jet fire, and explosion. Therefore Safety measures are necessary for this hazard. In addition, risks in semi-closed spaces such as tunnels, underground roads, and underground parking lots should be analyzed. In this study, an explosion experiment was conducted on a hydrogen tank used in a hydrogen vehicle to analyze the risk of a hydrogen vehicle explosion accident that may occur in a semi-closed space. As results, the effect on the structure and the human body was analyzed using the overpressure and impulse values for each distance generated during the explosion.

Last year, the Autonomous Vehicle Act was enacted to respond to deployment of autonomous vehicles. But the Act stipulates the operation of autonomous vehicle pilot zones, In addition, in order to analyze autonomous vehicle accidents and establish a reasonable damage compensation system, the Automobile Damage Compensation Guarantee Act was revised. But, It is necessary to seek plans for institutional development such as detailed concepts of self-driving cars and driving, a security certification system for securing safety of autonomous cooperative driving, and enhancement of the effectiveness of special cases related to personal information processing. I would like to seek ways to improve the legal system to respond reasonably to the deployment of autonomous vehicles.

Recently hydrogen fuel cell buses have been deployed for the public transportations. In order to introduce buses fueled by hydrogen successfully, the research results of hydrogen bus safety should be discussed and investigated significantly. Especially, Korean government drives research in terms of various applications of hydrogen energy to replace the conventional fuel energy resources and to improve the safety evaluation. Thus it is necessary to examine vehicle crashworthiness under side impact loadings. This study was focused on the simulation result evaluation of full bus model and simplified bus model with hydrogen fuel tank module and mounting system located below floor structure due to the significance of bus side impact accidents. The finite element models of hydrogen bus, fuel tank system and side impact moving barrier were set up and simulation results reported model performance and result comparison of two side impact models. Computational results and research discussion showed the conceptual side impact framework to evaluate hydrogen bus crashworthiness.

Vehicle occupant postures are anticipated to vary more widely during automated driving and to become more significant in terms of the autonomous vehicle safety. Experimental and computational approaches are needed to investigate and evaluate occupant behaviors during automated driving in general. However the validity and effect of such occupant postures are unknown, thus it is necessary to examine occupant behaviors and injury countermeasures for various occupant postures. This study was focused on the development and evaluation of restraint system model for occupant behavior examinations in the first step according to autonomous vehicle occupant safety. The finite element models of dummy and restraint system were set up and simulation results showed overall model performance and safety tolerances of different reclined occupant postures during frontal impact loading.

In this study, The behavior of an autonomous vehicle in an intersection accident situation is predicted. Based on a representative intersection accident situation from actual intersection accident database, simulation was performed by applying the automatic emergency braking algorithm used in the autonomous driving system. Accident reconstruction was performed based on the accident report of the representative accident situation. After applying the autonomous driving system to the accident-related vehicle, the tendency of intersection accidents that may occur in autonomous vehicles was identified and analyzed.

In this paper, an automated vehicle intersection collision accident was analyzed through simulation. Recently, the more automated vehicles are distributed, the more accidents related to automated vehicles occur. Accidents may show different trends depending on the sensor characteristics of the automated vehicle and the performance of the accident prevention system. Based on NASS-CDS (National Automotive Sampling System-Crashworthiness Data System) and TAAS (Traffic Accident Analysis System), four scenarios are derived and simulations are performed. Automated vehicles are applied with a virtual system consisting of an autonomous emergency braking system and algorithms that predict the route and avoid collisions. The simulations are conducted by changing the sensor angle, vehicle speed, the range of the sensor and vehicle speed range. A range of variables considered vehicle collision were derived from the simulation.

This paper present box feature estimation from LiDAR point cluster using maximum likelihood Method. Previous LiDAR tracking method for autonomous driving shows high accuracy about velocity and heading of point cluster. However, Assuming the average position of a point cluster as the vehicle position has a lower accuracy than ground truth. Therefore, the box feature estimation algorithm to improve position accuracy of autonomous driving perception consists of two procedures. Firstly, proposed algorithm calculates vehicle candidate position based on relative position of point cluster. Secondly, to reflect the features of the point cluster in estimation, the likelihood of the particle scattered around the candidate position is used. The proposed estimation method has been implemented in robot operating system (ROS) environment, and investigated via simulation and actual vehicle test. The test result show that proposed cluster position estimation enhances perception and path planning performance in autonomous driving.

This paper proposes actuator fault detection and adaptive fault-tolerant control algorithms using performance index and human-like learning for longitudinal autonomous vehicles. Conventional longitudinal controller for autonomous driving consists of supervisory, upper level and lower level controllers. In this paper, feedback control law and PID control algorithm have been used for upper level and lower level controllers, respectively. For actuator fault-tolerant control, adaptive rule has been designed using the gradient descent method with estimated coefficients. In order to adjust the control parameter used for determination of adaptation gain, human-like learning algorithm has been designed based on perceptron learning method using control errors and control parameter. It is designed that the learning algorithm determines current control parameter by saving it in memory and updating based on the cost function-based gradient descent method. Based on the updated control parameter, the longitudinal acceleration has been computed adaptively using feedback law for actuator fault-tolerant control. The finite window-based performance index has been designed for detection and evaluation of actuator performance degradation using control error.