A Learning-based Stochastic MPC Design for Cooperative Adaptive Cruise Control to Handle Interfering Vehicles
For developers of CACC systems, this work addresses the practical challenge of handling interfering vehicles, but the contribution is incremental as it combines existing methods (NN prediction, SMPC) in a straightforward way.
The paper proposes a learning-based stochastic MPC for CACC that uses a neural network to detect and predict cut-in maneuvers by interfering vehicles, then incorporates the cut-in probability into the controller to improve safety. Evaluated on realistic SPMD data, the approach enhances reaction to dangerous cut-ins.
Vehicle to Vehicle (V2V) communication has a great potential to improve reaction accuracy of different driver assistance systems in critical driving situations. Cooperative Adaptive Cruise Control (CACC), which is an automated application, provides drivers with extra benefits such as traffic throughput maximization and collision avoidance. CACC systems must be designed in a way that are sufficiently robust against all special maneuvers such as cutting-into the CACC platoons by interfering vehicles or hard braking by leading cars. To address this problem, a Neural- Network (NN)-based cut-in detection and trajectory prediction scheme is proposed in the first part of this paper. Next, a probabilistic framework is developed in which the cut-in probability is calculated based on the output of the mentioned cut-in prediction block. Finally, a specific Stochastic Model Predictive Controller (SMPC) is designed which incorporates this cut-in probability to enhance its reaction against the detected dangerous cut-in maneuver. The overall system is implemented and its performance is evaluated using realistic driving scenarios from Safety Pilot Model Deployment (SPMD).