Dr. Kelvin Cheu Speaks at Royal Institute of Technology in Stockholm
Dr. Kelvin Cheu, C2SMART’s associate director for research, presented a seminar on lane-changing research at the Royal Institute of Technology in Stockholm on June 13.
Recent Developments in Lane Changing Research
The two most frequent tasks performed by drivers are car-following and lane changing. Lane changing occurs less frequently but are more complex than car-following because the subject vehicle’s driver needs to consider the risk of collision with up to four surrounding vehicles. In car-following, the driver only responds to two vehicles (ahead and behind the subject vehicle). Due to the complexity in data collection, lane changing behavior has not studied as frequently as car-following until the recent years. In this research, a lane changing decision model has been developed based on the Fuzzy Inference Systems (FIS) using the well-known NGSIM data. The FIS was selected as it represents a human’s decision making process. First, a survey was first conducted to determine the important parameters used by most drivers in making lane changing decisions, and to assist in the construction of fuzzy sets and fuzzy membership functions. Given the inputs that describe the relative positions between the subject vehicle and its surrounding vehicles, the FIS fuzzifies the inputs, applies fuzzy rules, composes the outputs of the rules, and defuzzifies the composite output into a binary decision: “yes – change lane” or “no – do not change lane”. The proposed FIS lane changing model has been calibrated with NGSIM data collected at the I-80 Freeway and then tested with NGSIM data collected at U.S. Highway 101, both in California. The test results show that the FIS achieved 99.5% accuracy in making lane changing decisions (with reference to the U.S. Highway 101 NGSIM data). This accuracy is better than the TRANSMODELER’s lane changing decision model, and the models developed based on neural networks and support vector machines. The FIS model has the potential to be programmed into microscopic traffic simulation tools, lane change assist system in existing vehicles and even autonomous vehicles. The last part of the presentation reports the results of statistical tests, made possible by the NGSIM data, that validated the assumptions in most of the lane changing research.