基于车辆瞬间运动状态的1-M轨迹预测模型

In the applications of autonomous driving and V2X, vehicle trajectory prediction plays a crucial role.The current trajectory prediction frameworks based on deep learning all have strict requirements for the input data, which needs trajectory sequences with fixed lengths and intervals as inputs. This makes it difficult for the models to achieve the expected prediction results when facing incomplete or irregular trajectory data in practical applications.To address this issue, this paper proposes a 1to many trajectory prediction model based on the instantaneous motion state of the vehicle. This model encodes the instantaneous motion state of the vehicle combined with the road environment into a feature vector in a high-dimensional space and uses the Pearson correlation coefficient to establish the correlation between the feature vectors at adjacent moments of the same trajectory. Through a neural ordinary differential equation (ODE) solver to fits the derivative function of the feature vector with respect to time.The model can predict the vehicle\'s motion trajectory at any moment within the subsequent 1 second based on the vehicle trajectory data of one frame. The experimental results show that, on the CitySIM-Intersection A dataset, for the straight trajectory of this model, the average distance error(ADE) within 1 second is 0.16 meters and the final distance error (FDE) is 0.3 meters; for the curved trajectory, the ADE within 1 second is 0.21 meters and the FDE is 0.44 meters. It is an effective trajectory prediction model.