Building Minimal and Reusable Causal State Abstractions for Reinforcement Learning

Fuente: arXiv
Salvato in:
Dettagli Bibliografici
Autori principali: Wang, Zizhao, Wang, Caroline, Xiao, Xuesu, Zhu, Yuke, Stone, Peter
Natura: Preprint
Pubblicazione: 2024
Soggetti:
Accesso online:
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866915447716184064
author Wang, Zizhao
Wang, Caroline
Xiao, Xuesu
Zhu, Yuke
Stone, Peter
author_facet Wang, Zizhao
Wang, Caroline
Xiao, Xuesu
Zhu, Yuke
Stone, Peter
contents Two desiderata of reinforcement learning (RL) algorithms are the ability to learn from relatively little experience and the ability to learn policies that generalize to a range of problem specifications. In factored state spaces, one approach towards achieving both goals is to learn state abstractions, which only keep the necessary variables for learning the tasks at hand. This paper introduces Causal Bisimulation Modeling (CBM), a method that learns the causal relationships in the dynamics and reward functions for each task to derive a minimal, task-specific abstraction. CBM leverages and improves implicit modeling to train a high-fidelity causal dynamics model that can be reused for all tasks in the same environment. Empirical validation on manipulation environments and Deepmind Control Suite reveals that CBM's learned implicit dynamics models identify the underlying causal relationships and state abstractions more accurately than explicit ones. Furthermore, the derived state abstractions allow a task learner to achieve near-oracle levels of sample efficiency and outperform baselines on all tasks.
format Preprint
id arxiv_https___arxiv_org_abs_2401_12497
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle Building Minimal and Reusable Causal State Abstractions for Reinforcement Learning
Wang, Zizhao
Wang, Caroline
Xiao, Xuesu
Zhu, Yuke
Stone, Peter
Artificial Intelligence
Machine Learning
Robotics
I.2.9; I.2.8; I.2.6
Two desiderata of reinforcement learning (RL) algorithms are the ability to learn from relatively little experience and the ability to learn policies that generalize to a range of problem specifications. In factored state spaces, one approach towards achieving both goals is to learn state abstractions, which only keep the necessary variables for learning the tasks at hand. This paper introduces Causal Bisimulation Modeling (CBM), a method that learns the causal relationships in the dynamics and reward functions for each task to derive a minimal, task-specific abstraction. CBM leverages and improves implicit modeling to train a high-fidelity causal dynamics model that can be reused for all tasks in the same environment. Empirical validation on manipulation environments and Deepmind Control Suite reveals that CBM's learned implicit dynamics models identify the underlying causal relationships and state abstractions more accurately than explicit ones. Furthermore, the derived state abstractions allow a task learner to achieve near-oracle levels of sample efficiency and outperform baselines on all tasks.
title Building Minimal and Reusable Causal State Abstractions for Reinforcement Learning
topic Artificial Intelligence
Machine Learning
Robotics
I.2.9; I.2.8; I.2.6
url https://arxiv.org/abs/2401.12497