Research

Temporal-Spatial Transformer for Vehicle Trajectory Prediction in Intersection Scenario

[paper] (under review)

Jilin University, Oct. 2020~ present
Advisors: Prof. Sumin Zhang

• Designed an interactive agent selection strategy which is suitable for various types of intersection.
• Proposed a temporal-spatial transformer with a trainable reletive positional encoding block and a mask mechanism to effectively capture the features of vehicles’ trajectories and interactions and accurately predict vehicle future trajectory.
• Conducted extensive experiments in different trajectory prediction cases against different baselines to demonstrate the effectiveness of the proposed approach..

Attentive Spatio-Temporal Graph Deep Laerning for Multi-Modal Trajectory Prediction

[paper] (under review)

• Added the vectorized representations of each lane in the road map to the prediction model to realize road geometry-aware trajectory prediciton.
• Proposed a graph neural network based prediction model with both spatial and temporal attention mechanisms and map feature extraction block to take interactions and road geometry into account when predicting vehicle trajectories.