Visual navigation with multiple goals based on deep reinforcement learning

Learning to adapt to a series of different goals in visual navigation is challenging. In this work, we present a model-embedded actor-critic architecture for the multigoal visual navigation task. To enhance the task cooperation in multigoal learning, we introduce two new designs to the reinforcement...

Full description

Saved in:
Bibliographic Details
Main Authors: Rao, Zhenhuan, Wu, Yuechen, Yang, Zifei, Zhang, Wei, Lu, Shijian, Lu, Weizhi, Zha, ZhengJun
Other Authors: School of Computer Science and Engineering
Format: Article
Language:English
Published: 2022
Subjects:
Online Access:https://hdl.handle.net/10356/163231
Tags: Add Tag
No Tags, Be the first to tag this record!
Institution: Nanyang Technological University
Language: English
id sg-ntu-dr.10356-163231
record_format dspace
spelling sg-ntu-dr.10356-1632312022-11-29T05:03:31Z Visual navigation with multiple goals based on deep reinforcement learning Rao, Zhenhuan Wu, Yuechen Yang, Zifei Zhang, Wei Lu, Shijian Lu, Weizhi Zha, ZhengJun School of Computer Science and Engineering Engineering::Computer science and engineering Deep Reinforcement Learning Scene Generalization Learning to adapt to a series of different goals in visual navigation is challenging. In this work, we present a model-embedded actor-critic architecture for the multigoal visual navigation task. To enhance the task cooperation in multigoal learning, we introduce two new designs to the reinforcement learning scheme: inverse dynamics model (InvDM) and multigoal colearning (MgCl). Specifically, InvDM is proposed to capture the navigation-relevant association between state and goal and provide additional training signals to relieve the sparse reward issue. MgCl aims at improving the sample efficiency and supports the agent to learn from unintentional positive experiences. Besides, to further improve the scene generalization capability of the agent, we present an enhanced navigation model that consists of two self-supervised auxiliary task modules. The first module, which is named path closed-loop detection, helps to understand whether the state has been experienced. The second one, namely the state-target matching module, tries to figure out the difference between state and goal. Extensive results on the interactive platform AI2-THOR demonstrate that the agent trained with the proposed method converges faster than state-of-the-art methods while owning good generalization capability. The video demonstration is available at https://vsislab.github.io/mgvn. This work was supported in part by the National Key Research and Development Plan of China under Grant 2018AAA0102504; in part by the National Natural Science Foundation of China under Grant U1913204, Grant 61991411, and Grant U19B2038; in part by the Natural Science Foundation of Shandong Province for Distinguished Young Scholars under Grant ZR2020JQ29; and in part by the Shandong Major Scientific and Technological Innovation Project (MSTIP) under Grant 2018CXGC1503. 2022-11-29T05:03:31Z 2022-11-29T05:03:31Z 2021 Journal Article Rao, Z., Wu, Y., Yang, Z., Zhang, W., Lu, S., Lu, W. & Zha, Z. (2021). Visual navigation with multiple goals based on deep reinforcement learning. IEEE Transactions On Neural Networks and Learning Systems, 32(12), 5445-5455. https://dx.doi.org/10.1109/TNNLS.2021.3057424 2162-237X https://hdl.handle.net/10356/163231 10.1109/TNNLS.2021.3057424 33667168 2-s2.0-85102271666 12 32 5445 5455 en IEEE Transactions on Neural Networks and Learning Systems © 2021 IEEE. All rights reserved.
institution Nanyang Technological University
building NTU Library
continent Asia
country Singapore
Singapore
content_provider NTU Library
collection DR-NTU
language English
topic Engineering::Computer science and engineering
Deep Reinforcement Learning
Scene Generalization
spellingShingle Engineering::Computer science and engineering
Deep Reinforcement Learning
Scene Generalization
Rao, Zhenhuan
Wu, Yuechen
Yang, Zifei
Zhang, Wei
Lu, Shijian
Lu, Weizhi
Zha, ZhengJun
Visual navigation with multiple goals based on deep reinforcement learning
description Learning to adapt to a series of different goals in visual navigation is challenging. In this work, we present a model-embedded actor-critic architecture for the multigoal visual navigation task. To enhance the task cooperation in multigoal learning, we introduce two new designs to the reinforcement learning scheme: inverse dynamics model (InvDM) and multigoal colearning (MgCl). Specifically, InvDM is proposed to capture the navigation-relevant association between state and goal and provide additional training signals to relieve the sparse reward issue. MgCl aims at improving the sample efficiency and supports the agent to learn from unintentional positive experiences. Besides, to further improve the scene generalization capability of the agent, we present an enhanced navigation model that consists of two self-supervised auxiliary task modules. The first module, which is named path closed-loop detection, helps to understand whether the state has been experienced. The second one, namely the state-target matching module, tries to figure out the difference between state and goal. Extensive results on the interactive platform AI2-THOR demonstrate that the agent trained with the proposed method converges faster than state-of-the-art methods while owning good generalization capability. The video demonstration is available at https://vsislab.github.io/mgvn.
author2 School of Computer Science and Engineering
author_facet School of Computer Science and Engineering
Rao, Zhenhuan
Wu, Yuechen
Yang, Zifei
Zhang, Wei
Lu, Shijian
Lu, Weizhi
Zha, ZhengJun
format Article
author Rao, Zhenhuan
Wu, Yuechen
Yang, Zifei
Zhang, Wei
Lu, Shijian
Lu, Weizhi
Zha, ZhengJun
author_sort Rao, Zhenhuan
title Visual navigation with multiple goals based on deep reinforcement learning
title_short Visual navigation with multiple goals based on deep reinforcement learning
title_full Visual navigation with multiple goals based on deep reinforcement learning
title_fullStr Visual navigation with multiple goals based on deep reinforcement learning
title_full_unstemmed Visual navigation with multiple goals based on deep reinforcement learning
title_sort visual navigation with multiple goals based on deep reinforcement learning
publishDate 2022
url https://hdl.handle.net/10356/163231
_version_ 1751548586941743104