A spline-LSTM for Autonomous Truck Trajectory Prediction Based on Curve feature extraction

Complementing the missing trajectory data in real time is of great significance for the safe driving of autonomous trucks. This paper introduced a spline based long short-term memory network (LSTM) to tackle the positioning signal loss issues by realizing the real-time trajectory prediction of autonomous truck. The cubic B-spline curve is used to fit the trajectory data to obtain control points series. The LSTM is then used to train with historical data. As a result, without relying on the physical information of the road network and the surrounding environment information of the truck, truck trajectory prediction in 1–3 seconds is realized. A case study is conducted based on simulation data. Compared with other common methods, the proposed spline-LSTM shows advantages in both convergence speed and prediction accuracy, and is more suitable for real-time analysis.