Interaction-Aware Trajectory Prediction for Autonomous Vehicle Based on LSTM-MLP Model
Paper i proceeding, 2023

Trajectory prediction is one of the core functions of the autonomous vehicle, it greatly affects the rationality and safety of the decision-making module and the planning module. This is challenging because the motion of the target vehicle is affected by the interactive behavior of its surrounding vehicles. In this paper, we propose the interaction-aware trajectory prediction model for autonomous vehicles based on LSTM-MLP model. The encoder module encoded the history trajectories to extract the dynamic feature of each vehicle in the scenarios by the LSTM model, and then the interaction module captures the interactive feature using the MLP-Max Pooling model. In the end, the decoder module decodes the fusion feature which consists of the dynamic feature of the target vehicle and the interactive feature to output the future trajectory based on the LSTM model. The experiments are carried out on the publicly available NGSIM dataset, and the results show that the proposed model outperforms prior works in terms of RMSE value.

Trajectory prediction

LSTM-MLP model

Autonomous vehicle

Interaction aware


Zhiwei Meng

Jilin University

Chalmers, Arkitektur och samhällsbyggnadsteknik

Jiaming Wu

Chalmers, Arkitektur och samhällsbyggnadsteknik, Geologi och geoteknik

Sumin Zhang

Jilin University

Rui He

Jilin University

Bing Ge

Chinese Academy of Sciences

Smart Innovation, Systems and Technologies

2190-3018 (ISSN) 2190-3026 (eISSN)

Vol. 356 91-99
9789819932832 (ISBN)

6th KES International Symposium on Smart Transportation Systems, KES STS 2023
Rome, Italy,

Säkerhetstestning av intelligenta uppkopplade fordon i mixade trafikmiljöer

VINNOVA (Vinnova2019-03418), 2020-08-01 -- 2023-08-31.



Robotteknik och automation




Mer information

Senast uppdaterat