Industry-Academic cooperation project with Hyundai Motor Group
Department of Automotive Convergence
In the 4th industrial revolution, future autonomous vehicle technology is the most important technology. Future automotive technologies such as self-driving, connected cars, and eco-friendly vehicle are rapidly evolving. and electronic systems in automobiles are also increasing. The Hyundai Automotive Group is engaged in research and development of advanced next-generation convergence technologies based on technology in automobile related mechanical, electronic, and computer fields by operating the contract department, which is an industry-university cooperation project with Korea University. The MFR Laboratory is a laboratory belonging to the Department of Automotive Convergence. MFR Laboratory is working on ADAS (Advanced Driver Assistance Systems), a key technology for autonomous vehicles.
Random Traffic model reflecting Driver propensity
Research Territories
Automated Vehicle (AV) must be thoroughly evaluated before their release and deployment. A widely used evaluation approach is the Naturalistic-Field Operation Test (N-FOT) and Pre-Defined model simulation. N-FOT is essential, but there is a physical limitation of time and cost to evaluate all test cases in the scenario. And in the case of Pre-Defined Model Simulation, AV can “cheat” to pass predefined tests. In this paper, we propose a new model-based simulation framework that complements the limitations of existing evaluation methods. To evaluate autonomous vehicle, multi-driver traffic models that reflect driver propensity are designed based on Adaptive Artificial Potential Field(A-APF) and Model Predictive Control(MPC). The traffic behaviors of “HVs” as the major disturbance to AVs are derived using A-APF. All relevant objects in the driving scene can first be modeled separately in individual potential fields. Thereafter, these subfields can be normalized and weighted with Driver Propensity(DP) parameters, and are finally merged in an over-all representation. Because APF-based Path planning vehicles are recognized as a point mass, vehicle dynamics should be reflected. Therefore, this study proposes a 3DOF Lateral Vehicle Dynamic model to provide a suitable path for a non-holonomic system. In addition, appropriate control techniques are designed and compared with those models. This traffic model can be used for defining system requirements, system verification, system tuning or system sensitivity analysis to AV. The main adavantages of this evaluation model are higher variation of scenarios than N-FOT or other Model In the Loop Simulation(MILS) methods and repeatability.
Development of Longitudinal control system
using pedal information for heavy duty vehicle platoon
Research Territories
This study presents the design of longitudinal control system for the operation of automated heavy duty vehicles in platoons. Most recent studies mainly used the information received from the lead vehicle through communication and the information measured by the sensor. However, in the case of the heavy duty vehicles, there are a parasitic time delays such as actuators, sensors, and transport delay, and it is difficult to immediately reflect the driver's intention to drive. This study proposes a design of longitudinal control system using pedal information to overcome the time delay problem of heavy duty vehicles. For lead vehicle, this study estimated the acceleration through a nonlinear model for vehicle dynamics and for the following vehicles, the exact linearization method was used to linearize and normalize the input-output behavior of each vehicle. The simulated platoon consisted of five heavy duty vehicles of different weights, reflecting the characteristic of the heavy duty vehicles being weighted by cargo. The developed control system was tested on a simulation under driver pedal input conditions of gradual acceleration, gradual deceleration, and rapid deceleration. The string stability of the control system was confirmed by measuring the spacing error between successive vehicles.
Automated Parking Path Planning in Various Parking Environments Based on Algorithm Performance Evaluation
Research Territories
Autonomous vehicles recognize driving situation and can make decisions where to drive. The technology of autonomous vehicles can be categorized into three: cognition, decision, and control. In order to become a fully autonomous driving car, a function to park without a driver is necessary. it is called Autonomous Valet Parking (AVP). For the AVP, it is very important to create a path that does not collide with parked vehicles. The parking environments were divided into 9 scenarios according to the shape of the parking pattern, forward or reverse parking, and the location of obstacles. By using Graph search method and Dynamic programming, generate a parking path for 10 parking situations. In order for safety, speed, comfortability, weights were applied to the parking variables. As a result, the most suitable parking path planning method was founded for each situation.
L-Shape Feature and Feature Grouping Based Object Detection and SLAM for Autonomous Parking
Research Territories
Autonomous valet parking technology requires three functions: recognition, decision, and control. Of these, it is essential to recognize the location of autonomous vehicles and the parking space. A typical method of recognizing the position of a vehicle is to use GPS. However, in indoor environments such as indoor parking, GPS signals are not received, so other localization techniques should be used. SLAM (Simultaneous Localization and Mapping) is the simultaneous recognition of its location and surrounding environment. In order to implement SLAM, the environment is recognized by using attached sensor of the vehicle. Laser scanner is used in many autonomous vehicles because it provides more precise distance information than other sensors. The shape of the raw data representing the detected object depends on the position relative to the laser scanner. In a situation where the vehicle is covered with a neighboring parked vehicle and pillar, such as a parking lot, the shape of the vehicle detected by the laser scanner varies depending on the position of the laser scanner. Therefore, it is difficult to recognize a parked vehicle. Accordingly, there is a great difficulty in recognizing the position of the ego-vehicle and recognizing the parking space. In this paper, we propose an object detection algorithm based on L-shape feature and feature grouping and a SLAM technique using detected object in order to reduce the error of localization and mapping caused by the situation of vehicle detected due to laser scanner. The performance of the proposed algorithm was evaluated by experiments in an actual parking environment. We used the NAV350 from SICK to verify the performance of the algorithm. The reflector, which can be recognized by the NAV350 sensor, was attached to the vehicle and the performance was evaluated based on the position of the reflector. We compared localization errors with existing dual-weighted particle filters and detection rate with previous method.
