Cloud-enabled workflow-based real-time MPC for autonomous vehicle dynamical trajectory tracking

Model predictive control (MPC) has commonly been used in vehicle trajectory tracking problems, as it allows for predicting the future behaviors of multiple horizons and provides a certain degree of inherent robustness. However, due to its large computational amount, the real-time performance of MPC controller has been affected when more computing resources are required for complex vehicle dynamical models. The cloud workflow method has been proven effective for accelerating complex algorithms, such as deep learning, genetic calculation, etc., by adopting the distributed structure of cloud computing. This inspires the application of the cloud workflow method to data-intensive control problems. In this paper, a novel workflow-based MPC approach is proposed to accelerate the traditional computation mode. Firstly, a dynamical trajectory tracking method using MPC based on the alternating direction method of multipliers (ADMM) algorithm is designed for online optimization. Later, a cloud workflow construction method of MPC is designed to make full use of the distributed resources in cloud computing. In the meanwhile, the convergence conditions of the constructed MPC cloud workflow are proved for the vehicle dynamical trajectory tracking. Finally, based on a containerised workflow-based cloud control platform, we verify that the computational delays and average trajectory tracking errors are reduced by at most 68.1% and 87.6%, respectively.