JSAI2020

Presentation information

Cancelled

General Session

General Session » J-7 Agents

[1P4-GS-7] Agents: Game and collaboration

Tue. Jun 9, 2020 3:20 PM - 5:00 PM Room P (jsai2020online-16)

座長:沖本天太(神戸大学)

4:40 PM - 5:00 PM

[1P4-GS-7-05] Collaborative Planning with Implicit Guidance via Behavior

〇Ryo Nakahashi1, Seiji Yamada2,1 (1. SOKENDAI, the Graduate University for Advanced Studies, 2. National Institute of Informatics)

Keywords:Human Agent Interaction, Collaborative Planning, Theory of Mind

We consider the task in which an autonomous agent and a human cooperate in one task. In this task, since the ability of a human to plan is limited, it is generally better for an agent with a lot of computational resources to plan the action and guide the action of the human. However, if an agent explicitly guides a human's behavior one by one, the human does not feel the initiative of the task, and there is a danger of lowering the impression on the agent. We focused on an approach in which human behavior is implicitly guided by agent behavior. The agent calculates the optimal plan and then makes the action which makes a person easy to guess the optimal action based on the ``Theory of mind``, which, a cognitive mechanism in which humans infer the intention from the actions of others. As a result, we expect that humans will be voluntarily guided to better plans while maintaining their initiative for task. We have implemented an agent that performs implicit guidance by combining a model of bounded cognitive rationality with the conventional collaborative planning algorithm. In addition, through participant experiments in which participants take a simple collaborative task with an autonomous agent, we confirmed that implicit guidance improves the performance of a cooperative task while keeping the human initiative.

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