This paper proposes a learning method for adjusting the membership functions of antecedent fuzzy sets of each fuzzy rule in order to construct a fuzzy rule-based classification system. In the proposed method, three parameters that specify a non-symmetric triangular membership function are adjusted by an error-correction-based learning scheme. The grade of certainty of each fuzzy rule is also adjusted together with the membership functions of its antecedent fuzzy sets. The proposed method is illustrated by applying it to a simple two-dimensional pattern classification problem. Moreover, the effectiveness of the proposed method is demonstrated by computer simulations on some multi-dimensional pattern classification problems. Finally, an additional learning method for the fine tuning of classification boundaries between different classes is discussed.
This paper studies position/force hybrid control of a robot manipulator to interact with its uncertain flexible object. Because of its flexibility, the object dynamics will influence the robot's control system, and since it is a distributed parameter system, the object dynamics as seen from the robot's end-effector will change when the robot moves on its different locations. In this paper, the variation of the object dynamics as seen from the robot end-effector is formulated as an uncertain linear parameter-varying system (ULPV). Gain scheduled control is developed for controlling this kind of uncertain system. It is found, although the robot's position control loop is not influenced by the contact force, the robot's moving velocity influences the solution of the force control system's Riccati equation, which was used in constructing the system's force control input. Therefore, the position control loop should be designed with respect to the force control loop. Computer simulations show the effectiveness of our control approach.
An asynchronous modification of the block-parallel algorithm is presented for quadratic programming problems with separable objective functions. In particular, an implementation strategy is proposed to effectively decentralize the communication overhead associated with a specific processor of a parallel computer. Through extensive numerical experiments on the Connection Machine CM-5, the efficiency of the proposed method is examined for large scale problems with some block structure.
This paper is concerned with robust deadbeat control of sampled-data systems. The robustness in this paper is measured with L2- and L∞-induced norms, which are known to be useful for measuring robustness of sampled-data systems. The problems considered here are to find deadbeat controllers which minimize L2- and L∞-induced norms while achieving deadbeat settling with prescribed settling steps. It is shown that the problems can be reduced to some convex problems which can be solved numerically. A simple example is also presented to justify the usefulness of the proposed method