Search Results

Intelligent driving assistant systems have been studied meticulously for autonomous driving. When the systems have the responsibility for driving itself, such as in an autonomous driving system, it should be aware of its’ surroundings including moving vehicles and must be able to evaluate collision risk for the ego vehicle's planned motion. However, when recognizing surrounding vehicles using a sensor, the measured information has uncertainty because of many reasons, such as noise and resolution. Many previous studies evaluated the collision risk based on the probabilistic theorem which the noise is modeled as a probability density function. However, the previous probabilistic solutions could not assess the collision risk and predict the motion of surrounding vehicles at the same time even though the motion is possible to be changed by the estimated collision risk.

The majority of road accident is due to human error. Advanced Driver Assistance System (ADAS) has the potential to reduce human error and improve driving safety. Customers have shown a growing acceptance for ADAS technology. With the rising demand for safety and comfortable driving experience, the global market for ADAS is expected to grow to $67 billion by 2025. A reliable ADAS system requires an accurate and robust object-detection system. There is often a trade-off in tuning the system. On one hand, miss-detection can cause accidents; on the other hand, false-detection can result in ghost-braking and harm the driving experience. The ADAS system can access various information from different sources. However, a unified confidence model, which combines different indicators, has not been much studied in the literature. In this paper, we propose a data-driven method, which utilizes the features from radar, camera and the tracking system to produce a high-level confidence model.

Among the many sources of uncertainty in an unsteady computational fluid dynamics (CFD) simulation, the statistical uncertainty in the mean value of a fluctuating quantity (for example, the drag coefficient) is of practical importance for vehicle design and development. This uncertainty can be reduced by extending the simulation run length, however, this increases the computational cost and leads to longer turnaround times. Moreover, it is desirable to be able to run an unsteady CFD simulation for the minimum amount of time necessary to reach an acceptable amount of uncertainty in the quantity of interest. This work assesses several methods for calculating the uncertainty in the mean of an unsteady signal. Simulated noise is used to validate the methods, and evaluation is carried out using signals from CFD simulations of realistic vehicle geometries. Calculating the uncertainty in the difference between two signals is also discussed.

This paper examines the effect of rear effective backlight angle on vehicle contamination using contamination simulation results of a commercial vehicle. Highly-resolved time accurate computational fluid dynamics simulations were performed using a commercial Lattice-Boltzmann solver, to compare the rear end contamination with five different rear effective backlight angles. Additional aerodynamics simulations presented good correlation with published experimental data. The contamination results were compared with the aerodynamics simulation results in order to find trends between the two simulation types for different effective backlight angles.

A Light Detection And Ranging (LiDAR) is now becoming an essential sensor for an autonomous vehicle. The LiDAR provides the surrounding environment information of the vehicle in the form of a point cloud. A decision-making system of the autonomous car is able to determine a safe and comfort maneuver by utilizing the detected LiDAR point cloud. The LiDAR points on the cloud are classified as dynamic or static class depending on the movement of the object being detected. If the movement class (dynamic or static) of detected points can be provided by LiDAR, the decision-making system is able to plan the appropriate motion of the autonomous vehicle according to the movement of the object. This paper proposes a real-time process to segment the motion states of LiDAR points. The basic principle of the classification algorithm is to classify the point-wise movement of a target point cloud through the other point clouds and sensor poses.

In order to solve such problems that the millimeter-wave radar is of large computation, poor robustness and low precision of the target tracking algorithm, this paper presents an algorithmic framework for millimeter-wave radar tracking of target-vehicles. The target measurement information outside the millimeter- wave radar detection range is eliminated by the data plausibility judgment method based on the millimeter-wave radar detection parameters. Target clustering is made using Manhattan distance, to eliminate clutter interference and cluster multiple target measurements into one. The data association is made by use of nearest neighbor to determine the correspondence between information received measured by the radar and the real target. The vehicle is the key detection target of the vehicle millimeter-wave radar during road driving.

This work presents a method for estimating human body orientation using a combination of convolutional neural network (CNN) and stereo camera in real time. The approach uses the CNN model to predict certain human body keypoints then transforms these points into a 3D space using the stereo vision system to estimate the body orientations. The CNN module is trained to estimate the shoulders, the neck and the nose positions, detecting of three points is required to confirm human detection and provided enough data to translate the points into 3D space.

Threat assessment (TA) method is vital in the decision-making process of intelligent vehicles (IVs), especially for ADAS systems. In the research of TA, the probabilistic threat assessment (PTA) method is acting an increasing role, which can reduce the uncertainties of driver’s maneuvers. However, the driver behavior model (DBM) used in present PTA methods was mainly constructed by limited data or simple functions, which is not entirely reasonable and may affect the performance of the TA process. This work aims to utilize crash data extracted from Event Data Recorder (EDR) to establish more accurate DBM and improve the current PTA method in rear-end situations. EDR data with responsive maneuvers were firstly collected, which were then employed to construct the initial DBM (I-DBM) model by using the multivariate Gaussian distribution (MGD) framework. Besides, the model was further subdivided into six parts by two important risk indicators, Time-to-collision (TTC) and velocity.

Third generation advanced high strength steels (3GAHSS) are increasingly used in automotive for light weighting and safety body structure components. However, high material strength usually introduces higher springback that affects the dimensional accuracy. The ability to accurately predict springback in simulations is very important to reduce time and cost in stamping tool and process design. In this work, tension and compression tests were performed and the results were implemented to generate Isotropic/Kinematic hardening (I/KH) material models on a 3GAHSS steel with 980 MPa minimum tensile strength. Systematic material model parametric studies and evaluations have been conducted. Case studies from full-scale industrial parts are provided and the predicted springback results are compared to the measured springback data. Key variables affecting the springback prediction accuracy are identified.

Design of sheet metal drawing processes requires accurate information about the distribution of restraining forces, which is usually accomplished by a set of drawbeads positioned along the perimeter of the die cavity. This study is targeting bringing together the results of finite element analysis and experimental data in order to understand the most critical factors influencing the restraining force. The experimental study of the restraining force was performed using drawbead simulator tool installed into a tensile testing machine. Based upon the experimental results, it was observed that the restraining force of the given drawbead configuration is dependent upon the depth of bead penetration, friction between the drawbead surfaces as well as the clearance between the flanges of the drawbead simulator. This clearance is often adjusted during stamping operations to increase or decrease material inflow into the die cavity without any modification in the die.

Deep drawing is a sheet metal forming process in which metal blank is radially drawn into a forming die by the mechanical action of the punch. Dimensional tolerances and their variations are important aspects of quality control issues in this forming operation. In this regard, the spring-back effect is an inherent phenomenon that directly affects the final dimensions of the part produced. This research work is focused on analysis and control of spring-back in deep drawing processes. It is mainly focused on design and implementation/simulation of control strategies to minimize that. In this regard, the impact of various process parameters such as lubrication, punch speed, punch and die nose radius, and blank holding force is studied through design of experiment methodology. In particular, this study is focused on the design and development of various control strategies to minimize spring back in this process. An experimental set up is designed and developed to facilitate this research.

This work aims to providing an improved fit to continuously describe tensile test behavior over arbitrary quasi-static regression fit techniques. The tensile test, commonly defined by elastic, transient, and exponential regions, is represented here by a continuous curve spanning from the unstrained state to the post uniform regions. Since the model is continuous, proportionality and yield points between regions are not defined. This continuous behavior is described by an exponential expression defined in the logarithmic stress-strain coordinate system, from which the model fit is determined. In this logarithmic scale, we found that the data is bound by segments of concave and/or convex curvatures which end approach asymptotically towards straight lines. The coordinates of the fit in the logarithmic scale are defined at the intersection of the asymptotes, and the material fits are found from the optimum regression fit.

Forecasting the motion of the leading vehicle is a critical task for connected autonomous vehicles as it provides an efficient way to model the leading-following vehicle behavior and analyze the interactions. In this study, a personalized time-series modeling approach for leading vehicle trajectory prediction considering different driving styles is proposed. The method enables a precise, personalized trajectory prediction for leading vehicles with limited inter-vehicle communication signals, such as vehicle speed, acceleration, space headway, and time headway of the front vehicles. Based on the learning nature of human beings that a human always tries to solve problems based on grouping and similar experience, three different driving styles are first recognized based on an unsupervised clustering with a Gaussian Mixture Model (GMM).

Developing lightweight, stiff and crash-resistant vehicle body structures requires a balance between part geometry and material properties. High strength materials suitable for crash resistance impose geometry limitations on depth of draw, radii and wall angles that reduce geometric efficiency. The introduction of 3rd generation Advanced High Strength Steels (AHSS) can potentially change the relationship between strength and geometry and enable simultaneous improvements in both. This paper will demonstrate applicability of 3rd generation AHSS with higher strength and ductility to replace the 780 MPa Dual Phase steel in a sill reinforcement on the current Jeep Cherokee. The focus will be on formability, beginning with virtual simulation and continuing through a demonstration run on the current production stamping tools and press.

This study evaluated the interactions between sheet steel, lubricant and measurement system under typical sheet forming conditions using a fixed draw bead simulator (DBS). Deep drawing quality mild steel substrates with bare (CR), electrogalvanized (EG) and hot dip galvanized (HDG) coatings were tested using a fixed DBS. Various lubricant conditions were targeted to evaluate the coefficient of friction (COF) of the substrate and lubricant combinations, with only rust preventative mill oil (dry-0 g/m2 and 1 g/m2), only forming pre-lube (dry-0 g/m2, 1 g/m2, and >6 g/m2), and a combination of two, where mixed lubrication cases, with incremental amounts of a pre-lube applied (0.5, 1.0, 1.5 and 2.0 g/m2) over an existing base of 1 g/m2 mill oil, were analyzed. The results showed some similarities as well as distinctive differences in the friction behavior between the bare material and the coatings.

Lane keeping assist (LKA) is an autonomous driving technique that enables vehicles to travel along a desired line of lanes by adjusting the front steering angle. Reinforcement learning (RL) is one kind of machine learning. Agents or machines are not told how to act but instead learn from interaction with the environment. It also frees us from coding complex policies manually. But it has not yet been successfully applied to autonomous driving. Two control strategies using different deep reinforcement learning (DRL) algorithms have been proposed and used in the lane keeping assist scenario in this paper. Deep Q-network (DQN) algorithm with discrete action space and deep deterministic policy gradient (DDPG) algorithm with continuous action space have been implemented, respectively. Based on MATLAB/Simulink, deep neural networks representing the control policy are designed. The environment as well as the vehicle dynamics are also modelled in Simulink.

The thermal deflection associated with the conventional die heat treating procedure usually requires extra die grinding process to fine-tune the die surface. Due to the size of the production die, the grinding is time consuming and is not cost effective. The goal of the study is to develop a new die heat treating process utilizing the flexible laser heat treatment, which could serve the same purpose as the conventional die heat treating and avoid the thermal deflection. The unique look of the developed zebra pattern laser heat treating process is defined as the Zebra Line. The heat-treating parameters and processes were developed and calibrated to produce the laser heat treating on laboratory size dies, which were subjected to the die wear test in the laboratory condition. The USS HDGI 980 XG3TM steel was selected to be carried out on the developmental dies in the cyclic bend die wear test due to its high strength and coating characteristic.

Prediction of vehicle velocity is important since it can realize improvements in the fuel economy/energy efficiency, drivability, and safety. Velocity prediction has been addressed in many publications. Several references considered deterministic and stochastic approaches such as Markov chain, autoregressive models, and artificial neural networks. There are numerous new sensor and signal technologies like vehicle-to-vehicle and vehicle-to-infrastructure communication that can be used to obtain inclusive datasets. Using these inclusive datasets of sensors in deep neural networks, high accuracy velocity predictions can be achieved. This research builds upon previous findings that Long Short-Term Memory (LSTM) deep neural networks provide low error velocity prediction. We developed an LSTM deep neural network that uses different groups of datasets collected in Fort Collins, Colorado.

Autonomous vehicles are waiting to address the global automotive mobility challenges through an intelligent smart transportation system, which includes advanced sensor-actuator configurations to control, navigate, and drive the vehicles. Multi-sensor data fusion from the key sensors such as camera, radar, and lidar is used to achieve the environmental perception for autonomous vehicles by capturing the various attributes of the environment. Cameras are the dominant sensors to achieve the perception by providing vision capability to vehicles. The direct interface of the cameras with the dynamic driving environment carries numerous attack surfaces on the camera. Blinding attacks on the cameras are one of the critical attacks with an intention to blind the cameras either fully or partially by projecting light into the cameras to hide the objects which results in failure in object detection.

In recent years, with increase in external connectivity (V2X, telematics, mobile projection, BYOD) the automobile is becoming a target of cyberattacks and intrusions. Any such intrusion reduces customer trust in connected cars and negatively impacts brand image (like the recent Jeep Cherokee hack). To protect against intrusion, several mechanisms are available. These range from a simple secure CAN to a specialized symbiote defense software. A few systems (e.g. V2X) implement detection of an intrusion (defined as a misbehaving entity). However, most of the mechanisms require a system-wide change which adds to the cost and negatively impacts the performance. In this paper, we are proposing a practical and scalable approach to intrusion detection. Some benefits of our approach include use of existing security mechanisms such as TrustZone® and watermarking with little or no impact on cost and performance. In addition, our approach is scalable and does not require any system-wide changes.