Skip to main navigation Skip to search Skip to main content

StARformer: Transformer with State-Action-Reward Representations for Visual Reinforcement Learning

  • Stony Brook University

Research output: Chapter in Book/Report/Conference proceedingConference contributionpeer-review

34 Scopus citations

Abstract

Reinforcement Learning (RL) can be considered as a sequence modeling task: given a sequence of past state-action-reward experiences, an agent predicts a sequence of next actions. In this work, we propose State-Action-Reward Transformer (StARformer) for visual RL, which explicitly models short-term state-action-reward representations (StAR-representations), essentially introducing a Markovian-like inductive bias to improve long-term modeling. Our approach first extracts StAR-representations by self-attending image state patches, action, and reward tokens within a short temporal window. These are then combined with pure image state representations—extracted as convolutional features, to perform self-attention over the whole sequence. Our experiments show that StARformer outperforms the state-of-the-art Transformer-based method on image-based Atari and DeepMind Control Suite benchmarks, in both offline-RL and imitation learning settings. StARformer is also more compliant with longer sequences of inputs. Our code is available at https://github.com/elicassion/StARformer.

Original languageEnglish
Title of host publicationComputer Vision – ECCV 2022 - 17th European Conference, Proceedings
EditorsShai Avidan, Gabriel Brostow, Moustapha Cissé, Giovanni Maria Farinella, Tal Hassner
PublisherSpringer Science and Business Media Deutschland GmbH
Pages462-479
Number of pages18
ISBN (Print)9783031198410
DOIs
StatePublished - 2022
Event17th European Conference on Computer Vision, ECCV 2022 - Tel Aviv, Israel
Duration: Oct 23 2022Oct 27 2022

Publication series

NameLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume13699 LNCS
ISSN (Print)0302-9743
ISSN (Electronic)1611-3349

Conference

Conference17th European Conference on Computer Vision, ECCV 2022
Country/TerritoryIsrael
CityTel Aviv
Period10/23/2210/27/22

Keywords

  • Reinforcement learning
  • Sequence modeling
  • Transformer

Fingerprint

Dive into the research topics of 'StARformer: Transformer with State-Action-Reward Representations for Visual Reinforcement Learning'. Together they form a unique fingerprint.

Cite this