日本建築学会計画系論文集
Online ISSN : 1881-8161
Print ISSN : 1340-4210
ISSN-L : 1340-4210
深層強化学習を用いた自律走行搬送ロボットの動作の生成手法に関する基礎研究
藤澤 拓弥石田 航星
著者情報
ジャーナル フリー

2025 年 90 巻 828 号 p. 295-306

詳細
抄録

Methods are devised from the development of operation as three aspects of the performance of autonomous transport robots required at construction sites: explorability, safety, and versatility.

1. Explorability was enhanced by curiosity-driven exploration with intrinsic rewards, which reduced the number of externally rewarded items, and generalizability design improved accuracy.

2. Safety was achieved by creating safe areas for the agent and objects in the environment, as well as by granting negative rewards.

3. Generalizability is affected by the relationship between the variety of learning environments and learning accuracy fit.

著者関連情報
© 2025, 日本建築学会
前の記事 次の記事
feedback
Top