25
xxxxxxxxxx
-------------------------------------------------------
Submission information and review deadline
Authors and title:
Lingxiao Wang*, Shuo Pang
Robotic Odor Source Localization via End-to-End Recurrent Deep
Reinforcement Learning
Type of submission: Contributed paper
Conference: 2023 IEEE/RSJ International Conference on Intelligent
Robots and Systems (IROS)
Submission number: 346
Review deadline: April 25, 2023
There are no attachment files to be downloaded for this submission
Abstract:
This article presents a new olfactory-based navigation
algorithm that guides a mobile robot to find odor sources
in unknown environments. The proposed navigation algorithm
takes onboard sensor measurements as inputs and calculates
robot heading commands that direct the search agent (i.e.,
a mobile robot) to trace odor plumes back to the odor
source. We modeled this plume tracing process as a
partially observable Markov decision process (POMDP) since
the robot cannot fully observe the search environment and
adapted the twin delayed deep deterministic policy gradient
(TD3) to train the search agent. The long short-term memory
(LSTM) neural networks are utilized to construct the actor
and critic networks in the TD3 algorithm to fit the
partially observable environment. We employed the
curriculum learning to train the search agent in a
realistic plume tracing simulator, where the turbulence of
simulated airflows gradually increased. The agent was
implemented in different airflow environments after the
training. Simulation results show that in both laminar and
turbulent flow environments, the proposed algorithm
achieved a higher success rate and shorter averaged search
time compared to the original TD3 (i.e., without LSTM) and
traditional olfactory-based navigation algorithms.
---------------------------------------------------------
The file of 1 manuscript is attached.
To download any attachment to the submission (Video Attachment, Letters
of Financial Support) log in at
https://ras.papercept.net/conferences/scripts/start.pl as Reviewer for
IROS 2023, using your PIN 351067 and password, and follow the link
"Download" for the paper for which you wish to download an attachment.
You may also follow the quick link below.
To accept or to decline this request please follow the quick link
https://ras.papercept.net/conferences/scripts/confirm.pl?401&********
If this link is broken across lines then you may need to copy the
entire link and paste it into your browser address window
To accept or to decline the request you may also log in as a reviewer
for IROS 2023 at https://ras.papercept.net/conferences/scripts/start.pl
using your PIN 351067 and password and follow the link "Confirm or
decline" for the request.
To submit the review log in as a reviewer for IROS 2023 at
https://ras.papercept.net/conferences/scripts/start.pl using your PIN
351067 and password, and follow the link "Review" for the request. You
may type in or paste in your comments to the editorial staff and the
author.
Thank you for submitting your review on time.
If you do not have your password then please follow the link
https://ras.papercept.net/conferences/scripts/pinwizard.pl to retrieve
it.
---------------------------------------------------------
Prof. Zeyang Xia
Shenzhen Institutes of Advanced Technology
Chinese Academy of Sciences
1068 Xueyuan Ave.
518055 Shenzhen
China
E-mail address: zy.xia@siat.ac.cn
xxxxxxxxxx
-------------------------------------------------------
Submission information and review deadline
Authors and title:
Lingxiao Wang*, Shuo Pang
Robotic Odor Source Localization via End-to-End Recurrent Deep
Reinforcement Learning
Type of submission: Contributed paper
Conference: 2023 IEEE/RSJ International Conference on Intelligent
Robots and Systems (IROS)
Submission number: 346
Review deadline: April 20, 2023
There are no attachment files to be downloaded for this submission
Abstract:
This article presents a new olfactory-based navigation
algorithm that guides a mobile robot to find odor sources
in unknown environments. The proposed navigation algorithm
takes onboard sensor measurements as inputs and calculates
robot heading commands that direct the search agent (i.e.,
a mobile robot) to trace odor plumes back to the odor
source. We modeled this plume tracing process as a
partially observable Markov decision process (POMDP) since
the robot cannot fully observe the search environment and
adapted the twin delayed deep deterministic policy gradient
(TD3) to train the search agent. The long short-term memory
(LSTM) neural networks are utilized to construct the actor
and critic networks in the TD3 algorithm to fit the
partially observable environment. We employed the
curriculum learning to train the search agent in a
realistic plume tracing simulator, where the turbulence of
simulated airflows gradually increased. The agent was
implemented in different airflow environments after the
training. Simulation results show that in both laminar and
turbulent flow environments, the proposed algorithm
achieved a higher success rate and shorter averaged search
time compared to the original TD3 (i.e., without LSTM) and
traditional olfactory-based navigation algorithms.
---------------------------------------------------------
The file of 1 manuscript is attached.
To download any attachment to the submission (Video Attachment, Letters
of Financial Support) log in at
https://ras.papercept.net/conferences/scripts/start.pl as Reviewer for
IROS 2023, using your PIN 351067 and password, and follow the link
"Download" for the paper for which you wish to download an attachment.
You may also follow the quick link below.
To accept or to decline this request please follow the quick link
https://ras.papercept.net/conferences/scripts/confirm.pl?401&********
If this link is broken across lines then you may need to copy the
entire link and paste it into your browser address window
To accept or to decline the request you may also log in as a reviewer
for IROS 2023 at https://ras.papercept.net/conferences/scripts/start.pl
using your PIN 351067 and password and follow the link "Confirm or
decline" for the request.
To submit the review log in as a reviewer for IROS 2023 at
https://ras.papercept.net/conferences/scripts/start.pl using your PIN
351067 and password, and follow the link "Review" for the request. You
may type in or paste in your comments to the editorial staff and the
author.
Thank you for submitting your review on time.
If you do not have your password then please follow the link
https://ras.papercept.net/conferences/scripts/pinwizard.pl to retrieve
it.
---------------------------------------------------------
Prof. Zeyang Xia
Shenzhen Institutes of Advanced Technology
Chinese Academy of Sciences
1068 Xueyuan Ave.
518055 Shenzhen
China
E-mail address: zy.xia@siat.ac.cn