A Detection and Mitigation System for Unintended Acceleration: An Integrated Hybrid Data-driven and Model-based Approach
MetadataShow full item record
This study presents an integrated hybrid data-driven and model-based approach to detecting abnormal driving conditions. Vehicle data (e.g., velocity and gas pedal position) and traffic data (e.g., positions and velocities of cars nearby) are proposed for use in the detection process. In this study, the abnormal driving condition mainly refers to unintended acceleration (UA), which is the unintended, unexpected, uncontrolled acceleration of a vehicle. It is often accompanied by an apparent loss of braking effectiveness. UA has become one of the most complained-about vehicle problems in recent history. The data-driven algorithm aims to use historical data to develop a model that describes the boundary between normal and abnormal vehicle behavior in the vehicle data space. At first, several detection models were created by analyzing historical vehicle data at specific moments such as acceleration peaks and gear shifting. After that, these models were incorporated into a detection system. The system decided if a UA event had occurred by sending real-time vehicle data to the models and comprehensively analyzing their diagnostic results. Besides the data-driven algorithm, a driver model-based approach is proposed. An adaptive and rational driver model based on game theory was developed for a human driver. It was combined with a vehicle model to predict future vehicle behavior. The differences between real driving behavior and predicted driving behavior were recorded and analyzed by the detection system. An unusually large difference indicated a high probability of an abnormal event. Both the data-driven approach and the model-based approach were tested in the Simulink/dSPACE environment. It allowed a human driver to use analog steering wheels and pedals to control a virtual vehicle in real time and made tests more realistic. Vehicle models and traffic models were created in dSPACE to study the influences of UA and ineffective brakes in various roadway driving situations. Test results show that the integrated system was capable of detecting UA in one second with high accuracy. Finally, a brake assist system was designed to cooperate with the detection system, which reduced the risk of accidents.
Yu, Hongtao (2016). A Detection and Mitigation System for Unintended Acceleration: An Integrated Hybrid Data-driven and Model-based Approach. Doctoral dissertation, Texas A & M University. Available electronically from