This paper deals with a formation control problem of networked multi-agent systems via distributed pattern decision. We assume that several formation patterns are given without a leader which decides what formation pattern the group of agents should form. Each agent has to individually decide a possible formation pattern by watching the configuration of his neighborhood. Our control objective is to achieve one of the given formation patterns as a result of the distributed pattern decisions. We propose a new objective function consisting of formation errors on the cliques of networks, and design a formation controller based on the gradient flow of this clique-based objective function. The effectiveness of the proposed method is illustrated by simulations.
A passive walker is a robot which can walk down a shallow slope without active control or energy input, being powered only by gravity. This paper proposes a control law that can stabilize the gait of a quasi-passive walker by manipulating torque at the hip joint. The motion of the quasi-passive walker is divided into two modes: one is a sinusoidal mode and the other a hyperbolic sinusoidal mode. The controller is designed with a servo system which forces the motion of the sinusoidal mode to track the reference input signal obtained from the phase-plane trajectory of the hyperbolic sinusoidal mode. The generated gait is quite natural, because the input of the servo system is made based on the system dynamics. The results of simulations have demonstrated the effectiveness of the proposed control law.
An observer based failure detection filter is introduced, where actuator/sensor failures are modeled as a disturbance. A sufficient condition for designing the filter and explicit formulas of the filter gains are presented. Then, the filter design is applied to active mass dampers (AMDs), where AMD actuator failures are considered. It is shown that the AMD model obtained by system identification satisfies the condition for designing the fault detection filters each of which utilizes either position of the AMD weights or acceleration of the floors. The failure detection ability of the filters is also demonstrated through experiments.