Study of Driving Volatility in Connected and Cooperative Vehicle Systems

University Of Tennessee Knoxville, Knoxville TN

Investigators

Abstract

Innovative connected vehicle solutions can address major societal problems of safety, mobility, energy, and air quality. This research is to study the role of wireless connectivity and how information from vehicle and transportation system infrastructure based sensors can be integrated, processed, and disseminated. A technical challenge is enabling drivers and/or autonomous vehicles to take advantage of new and intelligent technologies, and estimate how they will respond to new forms of information, and how feedback (early warnings and control assistance) can help improve performance. Especially important in this context is driving volatility, characterized by hard accelerations/braking, jerky movements, sharp lane changes or turns, and abnormally high speeds. Can these be mapped to a combination of local and global traffic states surrounding the vehicles? Are they related to how surrounding vehicles behave and are they related to socio-demographics of drivers? These questions will be answered by leveraging connectivity and cooperation between vehicles and civil infrastructure. The research will involve the development of analytic procedures for understanding driving volatility, and the use of data for generating information and driver feedback. The research will advance the multidisciplinary field of driver behavior and cyber-physical systems, broaden participation of underrepresented groups in research, and enhance engineering education. The objective of this research is to model computationally efficient algorithms for predicting driver actions and volatility using information about their prior behaviors combined with positions and motions obtained via wireless communications. A Markov Decision Process framework will be used to develop models that anticipate instantaneous driver maneuver decisions. Driver-specific estimates of rewards and penalties for available maneuver choices in driving situations will be learned using Bayesian Inverse Reinforcement Learning and gossip algorithms. The effect of social persuasions from other drivers in a connected and cooperative vehicle system environment will be quantified. Subsequently, systematic representation of individual driver policies, along with emergent cyber-physical system information will enable: 1) preemptive warnings and assists to drivers based on anticipated maneuvers that are potentially unsafe, 2) expanded spatio-temporal sphere of influence to slow down vehicles approaching an incident, reducing chances of cascading incidents, and 3) maneuver suggestions to drivers that are sequentially optimized for the entire group (or small clusters) of following vehicles in the event of an incident. The potential widespread adoption of the research findings will benefit society by advancing key methodologies at the boundaries of connected vehicles systems.