[18a-A410-1~9] FS.1 Focused Session "AI Electronics"

△ [18a-A410-9] Computing with a pneumatic artificial muscle: implementations of the length estimation and bifurcation embedding

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

A pneumatic artificial muscle (PAM) is a typical soft actuator that realizes expansion/contraction or bending dynamics by air pressurization. It has been shown that PAM sensory values, such as pressure and electric resistance, can be exploited as an information processing resource by physical reservoir computing (PRC), which is a machine-learning architecture using physical dynamics. This study shows two demonstrations of the PAM PRC. First, we show that PRC can outperform external machine learning networks in the PAM length estimation. Second, PRC can embed periodic dynamics, chaotic dynamics, and these bifurcation structures in the PAM by the closed-loop control.