Support vector machines (SVM) are known to result in a quadratic programming problem, that requires a large computational complexity. One way to reduce this requirement is to apply the idea of incremental learning to SVMs, where only a subset of given examples is stored in an iteration and the complexity is linear with respect to the number of examples on average. When an incremental SVM stores the set of support vectors, it has a lower performance than an SVM in batch mode since examples that are not support vectors when given but become ones later are neglected. This paper indirectly evaluates the degradation of the SVM performance through the size of admissible region under ‘disc approximation’ and verifies results using computer simulations.
This paper considers a control problem of constrained linear systems. By investigating the input-output relation of finite-horizon linear systems, we clarify pairs of input and output signals which describe the dominant dynamics of the system. A control method is developed based on the proposed decomposition and fundamental properties of the resulting system are discussed. The proposed method is illustrated with numerical simulations and experiments.
Robust tracking of a GNSS signal in a harsh environment such as a severe ionospheric scintillation is a challenge for the civil aviation. The use of an inertial sensor would improve the tracking performance since the Doppler frequency caused by aircraft dynamics could be compensated by the inertial measurements. In order to evaluate such an aiding, an INS aided GPS tracking loop is developed by using a software receiver, and a preliminary flight test is conducted. A navigation grade INS tightly-coupled with GPS as well as a low-cost MEMS INS loosely-coupled with GPS are installed to provide aiding information. In addition, two GPS front end units with different clock (TCXO and OCXO) are installed to collect digitized IF data. Off-line analyses during aircraft take-off show that the noise band width in tracking loop can be reduced to three hertz by the aiding. Also, the Doppler aiding by a low-cost MEMS INS shows a similar performance with the aiding by a navigation grade INS.
In this paper, we propose a universal control formula with local LQ performance for feedback linearizable systems. We show the inverse optimality and the gain margin of the proposed controller. The proposed controller provides an easy transformation method from a local LQ controller to the nonlinear inverse optimal controller. The effectiveness of the proposed method is confirmed by computer simulation and experiments of a magnetic levitation system.
In this paper, the relationship between the internal stabilizing mechanism of passive dynamic walking and linear optimal control is investigated. In contrast to existing results, it is shown that the stabilizing mechanism can be characterized as a special type of optimal control. After verifying the feasibility of such a formulation numerically, it is proved that the internal stabilizing feedback coincides with a cheap optimal control law.
We propose a novel directional antenna device that couples 2D guided microwaves with 3D radiated beams. It consists of a dielectric scatterer array aligned in two-dimensional periodicity on a planar surface-waveguide. Each scatterer radiates a portion of the incident guided wave in a location-dependent linear phase delay. Therefore, the directivity of the synthesized scattered wave is controllable by tuning the period of the scatterers. We derive a guideline to design the scatterer pattern for beamforming and verify it with numerical simulations. Since the proposed method works reversibly both on sending and receiving, microwave paths for wireless communication and sensing can be designed smartly.
Heavy traffic congestion often occurs at an intersection of highway traffic. For releasing this congestion, we compare “zipper” and “non-zipper” mergings by numerical simulations. The “zipper” merging is the alternating merging of vehicles on two lanes and achieved only by the local communication of vehicles before merging, while the “non-zipper” merging is the disorderly merging of vehicles. In simulations we use a stochastic cellular automaton model for traffic with a slow-to-start rule. Numerical results show that the travel time of the “zipper” merging is shorter (longer) than that of the “non-zipper” merging in the case of a strong (resp. weak) slow-to-start effect. Moreover, it is observed that this reversal is strongly affected by the ratio of vehicles going straight without changing lanes.
This paper proposes to extend given reference signals in the iterative learning control context so that the resultant signals have the same right half plane zero structure as the given plants. This extension enables preventing control inputs exponentially increasing. The extending signals can be synthesized by solving a linear equation. The effectiveness of the proposed method is examined by a numerical example.
In this paper, we provide a new effective tuning method to obtain the optimal parameter of the feed-forward controller in a two-degree-of-freedom (2DOF) control system for the purpose of achieving the desired response without using a mathematical model of a plant. The first author proposed “fictitious reference iterative tuning” (which is abbreviated to FRIT) as an effective method for the tuning of parameters of a controller by using only one-shot experimental data instead of using mathematical models of a plant. Here, we extend FRIT to the tuning of the feed-forward controller in a 2DOF control system and develop the off-line computation so as to be done by the least squares method. For these purposes, we introduce a new cost function consisting of the fictitious reference and the actual data (since we do not have to do iterative computation in the least squares method, we call the proposed method here as “fictitious reference tuning”). Since the new cost function is quadratically-parameterized, it is possible to analyze how far the obtained parameter is apart from the desired one. Thus, we then derive a pre-filter which is applied to the actual data so as to guarantee that the obtained parameter is close to the desired one. We also show that the proposed method is applicable to the case in which the initial experiment is performed in the conventional 1DOF control system. Finally, we illustrate experimental results in order to show the utility and the validity of the proposed method.
In this paper, we provide a new method for updating a mathematical model of a plant and a controller, simultaneously, by using only one-shot experimental data. Here, we propose a fictitious controller which is described by the following triple: a nominal model of the plant, an initial controller designed for the nominal model, and a parameterized model of the plant. In addition, we introduce a cost function which involves the fictitious controller, the nominal model, and the actual experimental data of the closed loop with the initial controller. Then, for this very reason of the minimization of the introduced cost function, we show that utilizing the fictitious reference iterative tuning, which was proposed by the authors, enables us to obtain both a more accurate model and a more desirable controller than the nominal model and the initial controller, respectively. We also give quantitative evaluation of the difference between the nominal model and the updated one, and that between the initial controller and the updated one. Finally, we give examples in order to illustrate the validity of the proposed method.
In this paper, a data-driven design method that gives a desirable loop-shape is proposed for a single input plant. This method is applicable to the tuning of a static feedback gain and also that of the output matrix of a dynamical controller by using a transient response of the plant. Constraints on the feedback gain are derived from the maximum sensitivity and complementary sensitivity conditions based on the unfalsified control idea. A solution is obtained by solving a linear matrix inequality, where the integral gain of the loop transfer function is maximized subject to the constraints. Usefulness is demonstrated by two numerical examples.
It is well known that plants with dead-time are difficult to controll by using traditional control methods. For this, some controllers with dead-time have been proposed for the systems with dead-time, e.g. Internal Model Control (IMC) and the Smith-method. However, these controllers cause another problem, i.e. it would be difficult to realize the dead-time component in controller because of the memory limit of micro control unit (MCU). The sampling time has to be large in applications to the plant with large dead-time when each data size is sufficiently kept. Hence, a trade-off between the sampling time and the maximum quantization error exists by the memory limit. In this paper, a design method of dynamic quantizers is proposed for achieving small quantization error for control systems in MCU. The effectiveness of the proposed method is shown by numerical examples. Moreover, the proposed method is applied to a temperature control of a heat plate. Since the input-output relation of the temperature control system can be written with dead-time, 2-DOF IMC is introduced for this system. It is verified that the output with the proposed quantizer approximates the desired output under the memory limit.
This paper introduces a multilevel road network model to speed up the optimal route calculation on large size road networks by preprocessing and pre-computing. The multilevel road network is constructed by separating a large network into several smaller sub-networks and pushing up the boundary nodes. The new road sections on the multilevel network are decided by pre-computing the shortest path between the boundary nodes of each sub-network in a hierarchical manner. To optimize the structure of the multilevel road network, an effective genetic algorithm based clustering method is proposed, which can reduce the number of boundary nodes and road sections between boundary nodes of each sub-network greatly. Experimental results show that the proposed approach significantly reduces the search space of the route calculation algorithm over existing methods.