Search

This paper proposes a collision detection system to detect whether ego and target vehicles collide when both vehicles change from their lanes to the same lane. Although it is essential to predict this kind of collision for the active safety system, there is little literature on the case study. This paper presents the collision detection method using a Convolution Neural Network (CNN) consisting of four classes to predict collision risk on multi-lanes road conditions. The CNN is formed on stacked Occupancy Grid Maps (OGMs) based on point cloud data of the LiDAR and Radar sensors with in-vehicle sensor data for spatio-temporal information between vehicles. Further, we apply the open set recognition concept to the network to consider a conservative collision detection. The experimental results show the feasibility of the proposed collision detection system and the conservative decision about the confusing situation.

In this paper, we propose an estimation of the accurate vehicle position using Interacting Multiple Model Extended Kalman Filter (IMM EKF) when road surface varies. Since the vehicle has different cornering stiffness as the road surface varies, it is difficult to accurately estimate the position of the vehicle. To resolve this problem, we present the IMM EKF considering each model of different roads to improve the estimation performance. From the numerical simulation using MATLAB/CARSIM, we observed that the performance of the proposed algorithm improves vehicle positioning performance.

The decision of a driver intention of the surrounding vehicle is one of the essential parts of autonomous driving. Notably, the intention of cut-in is significant for longitudinal motion control of an ego vehicle on highways. However, there is no method directly to measure other vehicles’ intentions. In the paper, using a hidden Markov model (HMM), we conducted a study on decision-making for the driver intention of the surrounding vehicle in a curved road. We also formulated a parameter estimation method for a transition probability to get a maximal expectation of the intention with the Baum-Welch algorithm. To validate the proposed scheme, we performed a computational simulation to determine the driver’s intention. It turns out that the proposed system provides the driver intention earlier than a method with fixed transition probability.

This study aims to propose how autonomous vehicle overtakes an obstacle with model predictive control and makes a lateral ride comfortability by optimizing a jerk in a lateral direction. In this paper, we conducted a case study on collision avoidance in a waypoint tracking situation. To avoid the obstacle, we proposed a desired lateral offset from the global route and used the offset as a reference for avoidance. Then, we defined a quadratic cost function considering the jerk to minimize a motion sickness during the avoidance maneuver. To verify the method, we executed a computational simulation using actual waypoint data. As a result, we observed that the proposed scheme has more comfortability for obstacle avoidance than the method without optimizing the jerk.

In this paper, we propose an algorithm compensating for an unmatched disturbance which is difficult to thoroughly remove. In order to compensate for the disturbance of an autonomous driving vehicle, we exploit Deep Neural Network(DNN). It is well-known that we cannot compute the control command that completely compensates for any unmatched disturbance. In vehicle lateral dynamics using a lane keeping system with a look-ahead distance, there always exists the disturbance caused by a curvature of road. Thus, it is inevitable that the lane keeping system only with a feedback control has steady-state error in the lane center on a curved road. To resolve this problem, we utilize the feedback control system with DNN. The proposed method is unique in compensating for the unmatched disturbance in lane keeping system. From a simulation study, we observed that the proposed method could reduce lateral tracking error up to a half on a curved road.

This paper presents a motion prediction scheme of object vehicles based on the dynamic occupancy grid map considering movement of the vehicles by applying a temporal flow and a cascaded algorithm for support vector machine (SVM). We divided occupancy grid map into two types of upper-level and lower level. The upper-level occupancy grid is used to predict motion that the object vehicle can move into the ego vehicle and the lower-level one is needed for decision using the SVM for sensor resolution. The presented algorithm was validated with a experimental data set and the overall accuracy of classification was obtained 90.42% from a confusion matrix.

This paper presents a path generation method using curve fitting based on Kalman filter of LED way-points and tracking control for autonomous in-door valet parking system. Generally, if the given way-points are either irregular or uncertain, it could generate a discontinuous path that the following vehicle cannot track. To cope with this problem, we propose methods generating virtual way-points and fitting a curve using Kalman filter. Then, we improve tracking performance by using linear-quadratic regulator (LQR) controller. The presented algorithm was evaluated via simulation depicting in-door valet parking system and experiment using a test vehicle in in-door parking space equipped with LED way-points.

In this paper, we present a comparative study of two machine learning methods to estimate the road surface condition without directly estimating tire-road friction coefficient. It is well known that using either a vehicle model-based approach or an end-to-end artificial intelligent method is not satisfactory to estimate the tire-road friction coefficient due to sensor noise, parameter uncertainty, and disturbances. To cope with this problem, three feature vectors obtained based on the vehicle dynamics are utilized for support vector machine (SVM) and deep neural network (DNN) with a time-window approach. The effectiveness of the proposed method is verified using experimental data obtained with a test vehicle on proving grounds. From the experimental study, we observed that the road surface condition estimation using DNN is superior to that using SVM.

In this paper, we propose a lateral control (LC) method of an autonomous vehicle using an efficient model predictive control (MPC). Although model predictive control is effective in solving vehicle LC problems, it is difficult for a standard MPC to be applied to the lateral motion control of a vehicle due to its computational time required less than 10 [ms] for chassis control. Our proposed approximate explicit model predictive control (AEMPC) scheme does not require online quadratic programming but it uses pre-computed control gains. Therefore, it is suitable for chassis control requiring computation time less than a few msec. The proposed AEMPC is so simple and systematic that it can be implemented in solving the MPC problem in real-time with the digital signal processor of an electronic control unit (ECU) used for automotive applications. we validated how useful the proposed AEMPC is when applied to LC through computational experiment.

In this paper, we present discrete-time integral sliding model predictive control (DISMPC) as robust model predictive control (MPC) in presence of disturbances. It is common to use the dynamic lateral motion of a vehicle in designing lane keeping control with a look-ahead distance. However, this model includes an external signal by curved road and the look-ahead distance in lane keeping system (LKS) and assumes that lower controller of electric power steering (EPS) has ideal performance. The external signal and the unideal dynamic model of the EPS cause poor performance of the entire closed-loop system for lateral motion control. To resolve these problems, we use discrete-time integral sliding mode control (DISMC) as an inner loop controller while MPC is used in an outer control loop. The sliding mode surface is designed to minimize the effect of unmatched disturbances generated by the external signal for LKS. MPC is applied to improve the transient response performance when tracking a path. From simulation results on the curved road, we observed that the proposed method outperforms performance of MPC with the disturbance observer in LKS.

In this paper, we present a new object detection method in Adaptive Cruise Control (ACC) with the Support Vector Machine (SVM) algorithm using data from a radar system. ACC using Closest in Path Vehicle (CIPV) detects the object vehicle that comes in front of the vehicle's front radar. Therefore, if the object vehicle abruptly cuts into the lane ahead of ego vehicle, the speed of the ego vehicle should quickly reduce. This phenomenon makes passengers feel uncomfortable. To cope with this phenomenon, in this paper we propose multiple classifications of various driving situations using multi-class SVM. Classified data was used to detect the CIPV among nearby vehicles in ACC. The proposed method shows improved performance in predicting the motion of objects in advance over the conventional radar system so that it enable for the ACC system either to decelerate or to accelerate smoothly in advance. The performance of proposed method was validated via experimental results.

This paper introduces manufacturing of tilting pad gas bearing with a diameter of 5 mm and a length of 0.5∼1mm for power MEMS (Micro Electomechanical Systems) applications. Potential applications include micro gas turbines (100W) for mobile robots, turbo compressors/micro blower for micro fuel cells, micro bearings for pediatric heart, etc. Ultra fine gas bearing for micro component is so complicated that it is hard to fabricate by conventional machining from a manufacturing point of view. In the field of a manufacturing technology for micro parts, nanopowder processing has received a great deal of attention in recent years. In this study, the nanopowder tilting pad gas bearings were prepared by two different methods using: Ni-metal mold by LIGA process, and PDMS soft mold. The effect of the manufacturing condition on sintering properties of metal-nanopowder gas bearing was investigated. The density, shrinkage behavior and dimensional stabilities were analyzed, and the reproducibility was discussed in the compaction and sintering step.

This study is, thus, to suggest the way of sending safely messages for the Web-based service constituting EPC global network, of verifying whether the received messages are effective, of encoding the messages for safer sending and of certifying between systems, and then to implement the way.

This paper presents a novel inference technique based on the semantic primitives for the development of intelligent load distribution systems. We use the term "load distribution system" in this paper to designate switchgears, feeders and loads for large buildings and apartment complexes after the MOF. With the rapid development of the techniques in the fields of communications and computers, most electric facilities and protection relays have been digitalized. Extensive and detailed informations on the operating status are now available, which enable us to perform more sophisticated monitoring and management of the distribution loads. The operating status of the several local load distribution systems can be monitored even remotely by a central supervisory system. In this paper, we present an inference technique that can be used as one of the basic techniques in building intelligent power distribution panel systems. The technique can evaluate the distribution panel system operating status based on the semantic primitives, which are linguistic expressions in high abstraction levels. Results show that inference capabilities and thus the intelligence of the load distribution monitoring systems can be significantly enhanced with our approach. The proposed techniques are also being extended in developing an intelligent remote monitoring systems for large building and apartment complex load distribution systems.

In this paper, we propose an improved Maxshift method in Part 1 of JPEG2000. The proposed algorithm performs compression of an image that includes one or more ROI (region of interest) according to a demand of client using Maxshift method. We evaluate the performance of the proposed method with different bit rates. As a result of simulation, it shows that the proposed method increases PSNR vs. compression ratio performance and makes possible to transmit image with multiple ROIs over low bit rates above the Maxshift method.

We propose an improved Gaussian baseband filtered minimum shift keying (GMSK) quadrature modulator using a finite impulse response (FIR) filter whose coefficients are given by the difference of the phase response to remove the coefficient noise resulting from the conventional method. The modulator is designed by considering the effects resulting from the difference between the ideal analog signal and its digitized version in terms of the sampling noise, quantization noise, truncation noise, and coefficient noise. It is also confirmed by an experiment that it can be used for a digital cellular system.