Soft-bodied robots are attracting more and more attention for its potential in various applications in our living environment. Conventional analytical control, which has been optimized for rigid-bodied robots, cannot be used for soft-bodied robots because of the nonlinear mechanical response of soft materials. This forces robot designers to control them based on intuition. Therefore, we propose to apply ever-growing machine learning technologies to approximate the dynamics of soft material through force sensing, in order for the robots to acquire adaptive motions autonomously.
Matthew Ishige, Takuya Umedachi, Yoshihiro Kawahara