Pedestrian trajectory prediction by deep learning considering obstacles in the surrounding environment

In this paper, we propose a pedestrian trajectory prediction method for autonomous mobile robots. In many cases, there are many pedestrians in the environment in which the autonomous mobile robot runs. In such an environment, the robot needs to run safely for pedestrians. In order to avoid collision...

Full description

Bibliographic Details
Main Authors: Naoya SUGIURA, Takumi MATSUDA, Yoji KURODA
Format: Article
Language:Japanese
Published: The Japan Society of Mechanical Engineers 2021-06-01
Series:Nihon Kikai Gakkai ronbunshu
Subjects:
Online Access:https://www.jstage.jst.go.jp/article/transjsme/87/899/87_21-00125/_pdf/-char/en
_version_ 1811308635953823744
author Naoya SUGIURA
Takumi MATSUDA
Yoji KURODA
author_facet Naoya SUGIURA
Takumi MATSUDA
Yoji KURODA
author_sort Naoya SUGIURA
collection DOAJ
description In this paper, we propose a pedestrian trajectory prediction method for autonomous mobile robots. In many cases, there are many pedestrians in the environment in which the autonomous mobile robot runs. In such an environment, the robot needs to run safely for pedestrians. In order to avoid collisions with pedestrians and drive safely, it is important to predict future movements of pedestrians. In the conventional prediction method, the trajectory of a future pedestrian is often predicted from the position of a pedestrian in the past. However, in such cases, it is difficult to predict the movement to avoid obstacles such as walls and pillars around the pedestrian. In this study, point cloud data acquired by LiDAR is used to predict the behavior of pedestrians avoiding surrounding static obstacles. Based on the point cloud data, distances between the pedestrians and the static obstacles is calculated at each time. Then, input it into the network together with the transition of the pedestrian’s position to predict the future pedestrian’s position. In addition, we use attention mechanisms to model interactions between pedestrians. This makes predictions that consider not only static obstacles but also the effects of other pedestrians. The usefulness of this study is shown by performing accuracy evaluation using the dataset created in the simulation environment and the publicly available dataset.
first_indexed 2024-04-13T09:26:48Z
format Article
id doaj.art-88e9041acb624ae7acb491a7293cb1d9
institution Directory Open Access Journal
issn 2187-9761
language Japanese
last_indexed 2024-04-13T09:26:48Z
publishDate 2021-06-01
publisher The Japan Society of Mechanical Engineers
record_format Article
series Nihon Kikai Gakkai ronbunshu
spelling doaj.art-88e9041acb624ae7acb491a7293cb1d92022-12-22T02:52:24ZjpnThe Japan Society of Mechanical EngineersNihon Kikai Gakkai ronbunshu2187-97612021-06-018789921-0012521-0012510.1299/transjsme.21-00125transjsmePedestrian trajectory prediction by deep learning considering obstacles in the surrounding environmentNaoya SUGIURA0Takumi MATSUDA1Yoji KURODA2Graduate School of Science and Technology, Meiji UniversitySchool of Science and Technology, Meiji UniversitySchool of Science and Technology, Meiji UniversityIn this paper, we propose a pedestrian trajectory prediction method for autonomous mobile robots. In many cases, there are many pedestrians in the environment in which the autonomous mobile robot runs. In such an environment, the robot needs to run safely for pedestrians. In order to avoid collisions with pedestrians and drive safely, it is important to predict future movements of pedestrians. In the conventional prediction method, the trajectory of a future pedestrian is often predicted from the position of a pedestrian in the past. However, in such cases, it is difficult to predict the movement to avoid obstacles such as walls and pillars around the pedestrian. In this study, point cloud data acquired by LiDAR is used to predict the behavior of pedestrians avoiding surrounding static obstacles. Based on the point cloud data, distances between the pedestrians and the static obstacles is calculated at each time. Then, input it into the network together with the transition of the pedestrian’s position to predict the future pedestrian’s position. In addition, we use attention mechanisms to model interactions between pedestrians. This makes predictions that consider not only static obstacles but also the effects of other pedestrians. The usefulness of this study is shown by performing accuracy evaluation using the dataset created in the simulation environment and the publicly available dataset.https://www.jstage.jst.go.jp/article/transjsme/87/899/87_21-00125/_pdf/-char/entrajectory predictionmobile robotstatic obstacleattention mechanismspatial interaction
spellingShingle Naoya SUGIURA
Takumi MATSUDA
Yoji KURODA
Pedestrian trajectory prediction by deep learning considering obstacles in the surrounding environment
Nihon Kikai Gakkai ronbunshu
trajectory prediction
mobile robot
static obstacle
attention mechanism
spatial interaction
title Pedestrian trajectory prediction by deep learning considering obstacles in the surrounding environment
title_full Pedestrian trajectory prediction by deep learning considering obstacles in the surrounding environment
title_fullStr Pedestrian trajectory prediction by deep learning considering obstacles in the surrounding environment
title_full_unstemmed Pedestrian trajectory prediction by deep learning considering obstacles in the surrounding environment
title_short Pedestrian trajectory prediction by deep learning considering obstacles in the surrounding environment
title_sort pedestrian trajectory prediction by deep learning considering obstacles in the surrounding environment
topic trajectory prediction
mobile robot
static obstacle
attention mechanism
spatial interaction
url https://www.jstage.jst.go.jp/article/transjsme/87/899/87_21-00125/_pdf/-char/en
work_keys_str_mv AT naoyasugiura pedestriantrajectorypredictionbydeeplearningconsideringobstaclesinthesurroundingenvironment
AT takumimatsuda pedestriantrajectorypredictionbydeeplearningconsideringobstaclesinthesurroundingenvironment
AT yojikuroda pedestriantrajectorypredictionbydeeplearningconsideringobstaclesinthesurroundingenvironment