A deep learning optimized LQR method for enhanced formation control with embedded systems

Abstract

To achieve higher accuracy throughout the formation control processes and enhance precision in dynamic environments, particularly for the formation control of follower vehicles with embedded systems, this paper proposes a method and framework for vehicle formation control. An Ackermann-model based LQR controller is developed for lateral distance control and a PD controller for longitudinal distance control. To enhance the efficacy of the LQR controller, the Deep Deterministic Policy Gradient Derivative (DDPG) method is introduced into the control system. The DDPG networks are trained in a simulation environment and can subsequently predict LQR parameters in real-time experiments. The practical application of the methodology is showcased, and the concluding remarks emphasize the potential and superior performance of our proposed formation control approach by experimental comparison with other controllers. This methodology can be implemented in small vehicles that possess limited computational resources and is also suitable for scenarios requiring dynamic motion control with higher tracking accuracy and stability.