Intersection Crossing of Autonomous Vehicles for Communication Links with Packet Losses

This work aims at addressing the problem of intersection crossing for autonomous vehicles in the presence of a lossy communication channel (that may cause the loss of data packets in the communication) between the vehicle and a central coordinator. This work aims to provide a solution that guarantees the crossing of the intersection, despite the packet losses. Our approach consists of an optimal control algorithm in which at every time step the optimal control sequence simulating a communication ensemble between the vehicle and the coordinator is computed. Since the model is stochastic and the channel imperfect, the intersection crossing condition is modeled as a chance constraint. Once the control sequence is found, the first step is applied and the optimization is performed again over a shrunk horizon. A real-time state observer using optimal Kalman filter observation gains is used. The performance of our approach is tested on a simplified dynamic car model. The sensitivity of the generated control sequence with respect to some key parameters (such as failure probability allowed and packet drop probability) is also considered.