JSAI2023

Presentation information

Organized Session

Organized Session » OS-27

[2Q4-OS-27b] 強化学習の新展開

Wed. Jun 7, 2023 1:30 PM - 3:10 PM Room Q (601)

オーガナイザ:太田 宏之、甲野 佑、高橋 達二

1:30 PM - 1:50 PM

[2Q4-OS-27b-01] A unified control mechanism for action planning, execution, dialogue, and inference for the reward maximization

〇Yuuji Ichisugi1, Hidemoto Nakada1, Naoto Takahashi1, Izumi Takeuti1, Takashi Sano2 (1. AIST, 2. Toyo University)

[[Online]]

Keywords:hierarchical reinforcement learning, artificial general intelligence, planning, model-based reinforcement learning

We are developing an AI architecture that uses recursive reinforcement learning to control thought and behavior, in order to realize artificial general intelligence in the future.
Agents will act on the environment, interact with others, and reason about the state of the environment under unified control in order to maximize rewards.
In the future, we plan to implement a mechanism that allows agents to synthesize the control program based on their own experiences.
In this paper, we describe the overall architecture and propose a mechanism for action planning that works on top of it.
We implemented a prototype system of the proposed mechanism and verified its operation.

Authentication for paper PDF access

A password is required to view paper PDFs. If you are a registered participant, please log on the site from Participant Log In.
You could view the PDF with entering the PDF viewing password bellow.

Password