Structural Transformer Improves Speed-Accuracy Trade-Off in Interactive Trajectory Prediction of Multiple Surrounding Vehicles

Fast and accurate long-term trajectory prediction of surrounding vehicles (SVs) is critical to autonomous driving systems. In high-density traffic flows, strongly correlated vehicle behaviors require considering the interactions among multiple SVs when predicting their future trajectories. However, existing interactive prediction methods, most based on Long Short-Term Memory (LSTM), are suffering from slow prediction because they analyze SVs one by one and analyze trajectory sequence node by node. This paper presents a fast interactive trajectory prediction method called Structural Transformer which learns both spatial and temporal dependencies among multiple SVs in parallel. Specifically, our model first removes the internal states and loops of LSTM and replaces with a weighted self-reference mapping to realize parallel computation. Then, it embeds the relative spatial information of multiple SVs into trajectory states and reorganizes the self-reference mapping with neighbor-only interaction masks to achieve interactive prediction. Results on the NGSIM dataset show satisfyingly speed and accuracy performance on long-term trajectory prediction of multiple SVs. The longitudinal and lateral errors are reduced to 2.67m and 0.25m over 5s time horizon. The computational time of each step is only 12ms on a 2080ti GPU, which is over 4 times faster than the Structural LSTM.

I. Introduction

Autonomous driving is a convincing technology that can improve driving safety, reduce traffic congestion and release drivers’ driving burden [1]. An autonomous vehicle should be able to predict the traffic situation in the future and make appropriate decisions. However, the trajectories of surrounding road users are often difficult to predict due to the strong interactions with other road users [2]. This problem becomes more complicated when the number of prediction targets increases, especially in high-density traffic flows [3]. In addition, the observation for road users is usually imperfect and highly noisy due to the object occlusion and sensor uncertainty. Therefore, it is challenging to implement fast and accurate online trajectory prediction of multiple road users in complex traffic situations to achieve a higher level of autonomous driving.