Abstract
In order to deal with ultra multi degree-of-freedom (DOF) dynamical systems, we designed the dynamical system as an autonomous decentralized system, and controlled its behavior using a bottom-up scheme, i.e., by setting local rules for each DOF. And we set the local rule using neural network. We considered a simplified 242-DOF mass system, with subsystems represented by point masses connected together by linear actuators. We analyzed the behavior of this dynamical system by computer simulation. We applied the local rule to switch the P gain and the D gain of the actuators to realize the function: displacement controllability. This function was found to successfully emerge in the numerical simulation results. Hence, the effectiveness of the proposed design methodology is confirmed.
