[22p-A302-1~15] FS.1 Focused Session "AI Electronics"

1:00 PM - 1:15 PM JST(4:00 AM - 4:15 AM UTC)

[22p-A302-1] [Young Scientist Presentation Award Speech] Embedding bifurcation structures into a pneumatic artificial muscle using physical reservoir computing

〇Nozomi Akashi¹, Yasuo Kuniyoshi², Taketomo Jo³, Mitsuhiro Nishida³, Ryo Sakurai³, Yasumichi Wakao³, Kohei Nakajima² (1.Kyoto Univ., 2.Univ. Tokyo, 3.Bridgestone)

Keywords:

physical reservoir computing,pneumatic artificial muscle,soft robotics

Efficiently designing and learning diverse control patterns are crucial for adapting to various tasks and unknown environments in machine learning-based robot control. Recent research in machine learning has shown that learning the bifurcation structures of dynamical systems enables the generation of patterns qualitatively different from the training data. In this presentation, we demonstrate that a bifurcation structure containing periodic and chaotic patterns can be embedded into a soft actuator, which is called the McKibben-type pneumatic artificial muscle, using physical reservoir computing by training it to learn only one of the patterns. This result suggests that learning a specific pattern in robot control can lead to the simultaneous learning of various patterns associated with bifurcations.