Universal Value Iteration Networks: When Spatially-invariant is Not Universal

Li Zhang; Xin Li; Sen Chen; Hongyu Zang; Jie Huang; Mingzhong Wang

doi:10.1609/aaai.v34i04.6157

Back

Universal Value Iteration Networks: When Spatially-invariant is Not Universal

Conference paper

Peer reviewed

Universal Value Iteration Networks: When Spatially-invariant is Not Universal

Li Zhang, Xin Li, Sen Chen, Hongyu Zang, Jie Huang and Mingzhong Wang

Proceedings of the AAAI Conference on Artificial Intelligence, Vol.34(04), pp.6778-6785

Association for the Advancement of Artificial Intelligence (AAAI) Conference on Artificial Intelligence, 34th (New York, United States, 07-Feb-2020–12-Feb-2020)

Association for the Advancement of Artificial Intelligence

2020

DOI: https://doi.org/10.1609/aaai.v34i04.6157

Files and links (1)

url

https://doi.org/10.1609/aaai.v34i04.6157View

Published Version

Abstract

Knowledge Representation and Machine Learning

In this paper, we first formally define the problem set of spatially invariant Markov Decision Processes (MDPs), and show that Value Iteration Networks (VIN) and its extensions are computationally bounded to it due to the use of the convolution kernel. To generalize VIN to spatially variant MDPs, we propose Universal Value Iteration Networks (UVIN). In comparison with VIN, UVIN automatically learns a flexible but compact network structure to encode the transition dynamics of the problems and support the differentiable planning module. We evaluate UVIN with both spatially invariant and spatially variant tasks, including navigation in regular maze, chessboard maze, and Mars, and Minecraft item syntheses. Results show that UVIN can achieve similar performance as VIN and its extensions on spatially invariant tasks, and significantly outperforms other models on more general problems.

Details

Title: Universal Value Iteration Networks: When Spatially-invariant is Not Universal
Authors: Li Zhang (Author) - Beijing Institute of Technology
Xin Li (Author) - Beijing Institute of Technology
Sen Chen (Author) - Beijing Institute of Technology
Hongyu Zang (Author) - Beijing Institute of Technology
Jie Huang (Author) - Beijing Institute of Technology
Mingzhong Wang (Author) - University of the Sunshine Coast, Queensland, USC Business School - Legacy
Publication details: Proceedings of the AAAI Conference on Artificial Intelligence, Vol.34(04), pp.6778-6785
Conference details: Association for the Advancement of Artificial Intelligence (AAAI) Conference on Artificial Intelligence, 34th (New York, United States, 07-Feb-2020–12-Feb-2020)
Publisher: Association for the Advancement of Artificial Intelligence
Date published: 2020
DOI: 10.1609/aaai.v34i04.6157
ISSN: 2159-5399
Organisation Unit: USC Business School - Legacy; School of Science, Technology and Engineering; University of the Sunshine Coast, Queensland
Language: English
Record Identifier: 99492908902621
Output Type: Conference paper

Metrics

16 Record Views