Real-Time Trajectory Planning for Logistical Supply Transportation Using GRU Neural Networks

This paper focuses on the trajectory planning problem for automated ground vehicles in logistical supply transportation missions. Traditional optimization-based methods have high computational requirements and poor real-time performance. To better meet the real-time requirements of the mission, we propose a trajectory planning method based on the GRU neural network. Our method utilizes an optimization-based approach to generate a training dataset, and then employs the GRU model to learn the internal mapping relationship from state to control actions. This enables real-time control of the vehicle for path planning and obstacle avoidance. Additionally, our method introduces a low-cost strategy to augment the dataset by incorporating supplementary data into the training set, thereby enhancing the learning capability of the model. In our simulation experiments, our model demonstrates excellent path planning performance in various scenarios, with significantly reduced computation time. Moreover, compared to other neural network-based planning controllers, our approach exhibits enhanced competitiveness.