JSAI2020

Presentation information

Interactive Session

[4Rin1] Interactive 2

Fri. Jun 12, 2020 9:00 AM - 10:40 AM Room R01 (jsai2020online-2-33)

[4Rin1-19] Analysis of Process about Acquisition of Standing Skill in Agents with Different Physical Characteristics

Focusing on Evolutionary Learning with Neuroevolution

〇Masatoshi Yamada1, Ryosuke Takata2, Takafumi Sakamoto3, Yugo Takeuchi3 (1.Tokoha University, 2.Shizuoka University, 3.Graduate School of Science and Technology, Shizuoka University)

Keywords:Agent, Neuroevolution, Acquisition of Standing Skill, Evolutionary Learning

The purpose of this study is to analyze acquisition of standing skill process by three agents with different physical characteristics through evolutionary learning. For three agents learning, the method called NEAT (Neuro Evolution of Augmenting Topologies) was adopted. The higher the center of gravity of the agent's body, the higher rewards for each agents. As a result of this experiment, it was confirmed that Humanoid-v2 could not maintain standing skill, whereas Hopper-v2 and BipedalWalker-v2 had acquired the standing skill. From the discussion, it was suggested that (1) upper fixedness, (2) lower dynamic state, are related to the standing stability of agents.

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