Trajectory prediction model of airport flight area based on transformer network

Engineering notes

Key topics: autonomous driving, autonomous vehicle, trajectory prediction, prediction. See the paper for implementation details and experimental results.

Chinese explanation / 中文解读

中文解读待补充：本站会优先为端到端自动驾驶、BEV感知、3D目标检测、轨迹预测、路径规划、LiDAR感知等高价值论文补充中文说明。

Original abstract

A trajectory prediction network model based on airport road operation rules is proposed to ensure the safety of autonomous vehicles driving in the airport flight area. Firstly, the scene structure features of the airport flight area are extracted by the graph neural network, and the Transformer encoder encodes the historical motion data and high-definition map feature data of the traffic participants around the aircraft. Secondly, the temporal attention mechanism and the spatial attention mechanism are introduced to establish the spatio-temporal correlation of the encoded data and extract the spatio-temporal features. Combined with the rules and regulations of the airport aircraft activity area operation management manual, the rule hierarchy is constructed with the priority level, and the BERT-NER and Sentence-BERT are used to extract the rule features. Finally, the spatio-temporal features and rule features are aggregated into high-dimensional features, and the predicted trajectories of traffic participants are output by multiple cyclic decoding. The experimental results show that the prediction accuracy of the proposed model is improved by about 18.3 % compared with the baseline model, and the predicted trajectory can meet the rule constraints of the airport flight area.

Links and sources

Comments