Sorry, you need to enable JavaScript to visit this website.

facebooktwittermailshare

MEMORY-BASED PARAMETERIZED SKILLS LEARNING FOR MAPLESS VISUAL NAVIGATION

Abstract: 

The recently-proposed reinforcement learning for mapless visual navigation can generate an optimal policy for searching different targets. However, most state-of-the-art deep reinforcement learning (DRL) models depend on hard rewards to learn the optimal policy, which can lead to the lack of previous diverse experiences. Moreover, these pre-trained DRL models cannot generalize well to un-trained tasks. To overcome these problems above, in this paper, we propose a Memorybased Parameterized Skills Learning (MPSL) model for mapless visual navigation. The parameterized skills in our MPSL are learned to predict critic parameters for un-trained tasks in actor-critic reinforcement learning, which can be achieved by transferring memory sequence knowledge from long short term memory network. In order to generalize into un-trained tasks, MPSL aims to capture more discriminative features by using a scene-specific layer. Finally, experiment results on an indoor photographic simulation framework AI2THOR demonstrate the effectiveness of our proposed MPSL model, and the generalization ability to un-trained tasks.

Contributions:
1.Parameterizing previous knowledge facilitates generalization to un-trained tasks.
2.Policy parameters are changed with task parameters and entered targets.
3.Feature alignment strategy to learn distinguishable features between different states.

up
1 user has voted: yuyang liu

Paper Details

Authors:
Yuyang Liu, Yang Cong and Gan Sun
Submitted On:
11 September 2019 - 11:06pm
Short Link:
Type:
Poster
Event:
Presenter's Name:
Yuyang Liu
Paper Code:
1881
Document Year:
2019
Cite

Document Files

The poster of 1881.

(15)

Subscribe

[1] Yuyang Liu, Yang Cong and Gan Sun, "MEMORY-BASED PARAMETERIZED SKILLS LEARNING FOR MAPLESS VISUAL NAVIGATION", IEEE SigPort, 2019. [Online]. Available: http://sigport.org/4598. Accessed: Nov. 12, 2019.
@article{4598-19,
url = {http://sigport.org/4598},
author = {Yuyang Liu; Yang Cong and Gan Sun },
publisher = {IEEE SigPort},
title = {MEMORY-BASED PARAMETERIZED SKILLS LEARNING FOR MAPLESS VISUAL NAVIGATION},
year = {2019} }
TY - EJOUR
T1 - MEMORY-BASED PARAMETERIZED SKILLS LEARNING FOR MAPLESS VISUAL NAVIGATION
AU - Yuyang Liu; Yang Cong and Gan Sun
PY - 2019
PB - IEEE SigPort
UR - http://sigport.org/4598
ER -
Yuyang Liu, Yang Cong and Gan Sun. (2019). MEMORY-BASED PARAMETERIZED SKILLS LEARNING FOR MAPLESS VISUAL NAVIGATION. IEEE SigPort. http://sigport.org/4598
Yuyang Liu, Yang Cong and Gan Sun, 2019. MEMORY-BASED PARAMETERIZED SKILLS LEARNING FOR MAPLESS VISUAL NAVIGATION. Available at: http://sigport.org/4598.
Yuyang Liu, Yang Cong and Gan Sun. (2019). "MEMORY-BASED PARAMETERIZED SKILLS LEARNING FOR MAPLESS VISUAL NAVIGATION." Web.
1. Yuyang Liu, Yang Cong and Gan Sun. MEMORY-BASED PARAMETERIZED SKILLS LEARNING FOR MAPLESS VISUAL NAVIGATION [Internet]. IEEE SigPort; 2019. Available from : http://sigport.org/4598