A Shared Control Approach for Autonomous Vehicles via Driver Behaviors Learning

Vehicles today are in a phase between human control and full autonomy. A reliable and stable approach that shares the decisions of both human and autonomous controllers is needed. This paper introduces a shared control approach to offer control signal from autonomous controller to complement commands of drivers to reduce driving risks and improve driving performance. Knowing future actions of drivers is essential in final decision making. To predict the behavior of driver, Gaussian process regression is adopted to model the driver with the data gathered in the driving process. Utilizing the drivers model and vehicles model, autonomous signal is provided by Model Predictive Controller (MPC) and combined with commands of drivers in certain proportion constituting the final outputs. With the shared control algorithm, both tracking and heading errors are significantly eliminated.