Realization of Multi Contact Behaviors Using Contact-State Transition Model and Reinforcement Learning by Musculoskeletal Humanoid

A human-being often does contact-rich behaviors in daily lives. Contact-rich behaviors require human or robots to handle friction force between environment and their body. We propose contact-state transition model to achive contact-rich behaviors. The contact-state transition model represents dynamics of a robot and an environment which contacts with the robot. First, we confirm the effectiveness of contact-state transition model by conducting the hip shuffle experiment on the chair. Finally, we conbine contact-state transition model and reinforcement learning, and realize sit back motion in the musculoskeletal humanoid.