Home // ICAS 2023, The Nineteenth International Conference on Autonomic and Autonomous Systems // View article

Multi-Agent Planning Method Using Affordances from Environment

Authors:
Sawako Tajima
Daiki Takamura
Daiki Shimokawa
Reo Kobayashi
Reo Abe
Satoshi Kurihara

Keywords: multi-agent planning, action selection, affordance

Abstract:
For autonomous actors, such as robots to achieve their goals while operating adaptively in a dynamically changing environment, they need to accurately recognize changing conditions from moment to moment. Planning research for finding a sequence of actions to achieve a goal has a long history, and in recent years, machine learning has become the mainstream method for finding the optimal sequence of actions. However, it is difficult to deal with unexpected situations using this method, and in such cases, the real-time performance is lost due to the blind search for a sequence of actions that will achieve the goal. Living organisms, such as ourselves, have learned how to avoid blind search by perceiving environmental affordances. In this paper, we propose a method for robots to use affordances to recognize their situation accurately and to seek a sequence of actions to achieve their goals efficiently. Affordances are common sense in an individual situation, i.e., tacit knowledge, and conventionally can only be constructed manually, which has the limitation that they cannot be scaled. However, large-scale language models that have recently emerged may contain tacit knowledge, and we have developed a method for extracting this tacit knowledge. In this study, we incorporated this method into a multi-agent planning system that is highly adaptive to dynamic environmental changes. We confirmed that a sequence of actions can be efficiently obtained to achieve a goal by using affordances.

Pages: 19 to 24

Copyright: Copyright (c) IARIA, 2023

Publication date: March 13, 2023

Published in: conference

ISSN: 2308-3913

ISBN: 978-1-68558-053-7

Location: Barcelona, Spain

Dates: from March 13, 2023 to March 17, 2023