Applying map-masks to Trajectory Prediction for Interacting Traffic-Agents - ENSTA Paris - École nationale supérieure de techniques avancées Paris Accéder directement au contenu
Communication Dans Un Congrès Année : 2019

Applying map-masks to Trajectory Prediction for Interacting Traffic-Agents

Résumé

Autonomous vehicles perceive their surroundings and foresee the future behaviour of all other relevant interacting traffic-agents in order to navigate safely and to operate in the public road networks. Trajectory prediction is difficult due to the stochastic manner these different types of traffic-agents interact with each other and the changing navigation context. In this work, we propose a recurrent artificial neural network LSTM encoder-decoder based architecture to predict the movement of traffic agents. It uses map-masks of the area surrounding the ego vehicle and previous trajectory information to predict the trajectory of interacting traffic agents. This paper compares the proposed approach with LSTM baselines,using the NuScenes dataset which includes LiDAR point-cloud ground-truth data for traffic agents plus map information. Experimental results show that the proposed method outperforms the baselines based on the prediction accuracy.
Fichier principal
Vignette du fichier
TrajPred_Paper.pdf (437.64 Ko) Télécharger le fichier
Origine : Fichiers produits par l'(les) auteur(s)
Loading...

Dates et versions

hal-02393250 , version 1 (04-12-2019)

Identifiants

  • HAL Id : hal-02393250 , version 1

Citer

Vyshakh Palli Thazha, David Filliat, Javier Ibañez-Guzmán. Applying map-masks to Trajectory Prediction for Interacting Traffic-Agents. Deep Learning for Automated Driving (DLAD) workshop, IEEE International Conference on Intelligent Transportation Systems, Oct 2019, Auckland, New Zealand. ⟨hal-02393250⟩
125 Consultations
245 Téléchargements

Partager

Gmail Facebook X LinkedIn More