This project has addressed the formal modeling, analysis, and design of cyberphysical transportation systems with human-in-the loop. In particular, this project is based on the synergy of control theory, formal methods, and human factors research, targeted specifically to address safety issues in next-generation transportation systems. Next-generation transportation systems will feature a mixture of autonomous vehicles communicating with each other and with the surrounding infrastructure, vehicles that are only partially autonomous and keep the driver in the loop, which may or may not be connected to the communication infrastructure, and human-driven vehicles. In these situations, it is crucial to design the automated controllers on the autonomous vehicles such that they can guarantee collision-free routes. At the same time, it is necessary to design active safety systems on-board the partially autonomous vehicles such that they are able to predict human drivers’ decisions and intervene through warnings and overrides to keep safety. Such design tasks are a formidable challenge due to a number of factors, including the complexity of the system (many agents) and the uncertainty that characterizes the actions of human-driven vehicles and of drivers in general. This project focused on tackling these challenges in complex traffic scenarios, chiefly road intersections.

The main outcomes of the project are as follows. We developed provably safe collision avoidance algorithms for vehicles at large intersections, which are least conservative and can be implemented in real-time. We were able to overcome the computational complexity bottlenecks that plague most formal approaches to safety design and could achieve real-time computation by leveraging the monotone dynamical structure of traffic. Concurrently, we performed experiments with human subjects in our driving simulator in order to construct models of human driving behavior near intersections. These models take the form of hybrid automata that are partially ordered according to the level of uncertainty that each of the crucial parameters accounts for. These models were validated also in the driving simulator and were used to design provably safe active safety systems for partially automated vehicles. These active safety systems issue warnings and overrides when necessary and come with a quantified safety level. While safety control design for hybrid automata subject to stochastic uncertainty is in general a computationally daunting task, we were able to develop efficient algorithms for real-time implementation by leveraging the monotone dynamics of vehicle driving.  We therefore expect that our algorithms will be applicable to many future transportation systems, complementing data-driven approaches, for the safe and efficient operation of autonomous vehicles, human-driven vehicles, and partially automated vehicles all sharing our roads.

Last Modified: 01/21/2017
Modified by: Domitilla Del Vecchio