[2J5-GS-2] Machine learning: Advancement reinforcement learning (1)

[2J5-GS-2-04] Development of AMD vibration control system using reinforcement learning

〇Yuya Aoyama¹, Masayuki Yamanaka¹, Koichi Nakatsuka¹, Takahiro Suzuki¹, Takeshi Hata¹, Takeshi Sano¹, Taku Yoshioka², Atsushi Izawa², Tetsuo Shiihashi² (1. Obayashi Corporation, 2. Laboro.AI)

Keywords:Leinforcement Learning, AMD(Active Mass Damper), Vibration Control, Control Law, Floor Vibration

The purpose of this study was to obtain a high vibration control effect by learning a control law of AMD which reduces the vertical vibration of the floor etc. using reinforcement learning. The simulation using the learned control law and the experiment applied to the actual system confirmed that the vibration control effect exceeded the control law by the conventional optimal control.