2020
DOI: 10.1109/tro.2020.2994002
|View full text |Cite
|
Sign up to set email alerts
|

Towards Generalization in Target-Driven Visual Navigation by Using Deep Reinforcement Learning

Abstract: In this work, we address generalization in targetdriven visual navigation by proposing a novel architecture composed by two networks, both exclusively trained in simulation. The first one has the objective of exploring the environment, while the other one of locating the target. We test our agent in both simulated and real scenarios, and validate its capabilities through extensive experiments with previously unseen goals and unknown mazes, even much larger than the ones used for training.

Help me understand this report

Search citation statements

Order By: Relevance

Paper Sections

Select...
1
1
1

Citation Types

0
38
0

Year Published

2021
2021
2024
2024

Publication Types

Select...
6
1

Relationship

0
7

Authors

Journals

citations
Cited by 77 publications
(38 citation statements)
references
References 35 publications
(8 reference statements)
0
38
0
Order By: Relevance
“…In previous research [45] , the DRL algorithm was found to be generalizable to new scenarios, but at the expense of a decrease in performance and the need to fine-tune the network. To improve the generalization ability of the visual navigation algorithm, Devo et al [52] proposed the importance weighted actor-learner architecture, a new framework comprising object localization and navigation networks. The object localization network takes the target image and current frame as input, and outputs a six-dimensional vector that represents the position of the target in the current frame.…”
Section: Developmentmentioning
confidence: 99%
See 1 more Smart Citation
“…In previous research [45] , the DRL algorithm was found to be generalizable to new scenarios, but at the expense of a decrease in performance and the need to fine-tune the network. To improve the generalization ability of the visual navigation algorithm, Devo et al [52] proposed the importance weighted actor-learner architecture, a new framework comprising object localization and navigation networks. The object localization network takes the target image and current frame as input, and outputs a six-dimensional vector that represents the position of the target in the current frame.…”
Section: Developmentmentioning
confidence: 99%
“…In addition to reducing the state input size, Devo et al [52] designed a two-network architecture comprising an object localization network and a navigation network to solve the generalization problem. This architecture reduces the state-space dimension of the navigation network by preprocessing images through the object localization network.…”
Section: Solutionmentioning
confidence: 99%
“…Generalization across environments is discussed in [12]. The authors trained the agent in domain-randomized mazelike environments and experimented with a robot in a small maze.…”
Section: Related Workmentioning
confidence: 99%
“…• The simulator often provides the agent with features that are not available in the real world: the segmentation masks [4], [6], [10], distance to the goal, stopping signal [4]- [7], [11], [12], [14], etc. This information is given either as one of the agent's inputs [4], [10] or in the form of an auxiliary task [6].…”
mentioning
confidence: 99%
See 1 more Smart Citation