Deep neural network-based multi-stage autonomous guided vehicles trajectory optimization in material transfer systems

Abstract

This article primarily focuses on the design, testing, and validation of a deep neural network (DNN)-based control scheme for predicting multi-stage trajectory optimization in autonomous ground vehicles (AGVs) used in material transportation systems. The proposed design employs a multilayer structure. In the first layer, a desensitized trajectory optimization method is iteratively applied to establish a set of time-optimal multi-stage trajectories, taking into account noise-perturbed initial configurations. Subsequently, DNNs are trained using the preplanned optimal transportation trajectory dataset to learn the functional relationship between the system’s state and control actions in the second layer. To further enhance DNN performance, a simple yet effective data aggregation approach is designed and implemented. These trained DNNs are then utilized as motion controllers to generate real-time feedback actions. Numerical experiments are conducted to showcase the effectiveness and real-time applicability of the proposed control scheme in planning the multi-stage AGV transportation process.