A design method of state estimation observer under outliers and data-lost environment is proposed in this paper. In networked control systems, the data-lost is one of the most important topic for solving. When data-lost occurred, the output data for the control system is disappeared. The data-lost is significant problems if we want to use state feedback. To overcome negative impact by outliers and data-lost, we propose a method to disappear negative impact by outliers and data-lost using median information. The effectiveness of the proposed observer is evaluated by numerical examples.
3-D soft underwater robot using piezoelectric fiber composite is being developed by authors' group. The purpose of this report is to integrate driving systems such as battery, amplifier for piezoelectric fiber composite, microcomputer into the robot, so that the robot can swim in the water without any wiring. First, we design a small driving system with considering the basic requirements. Next, we design and make the prototype based on the structure of real trout in witch built-in driving system can be included. To confirm the basic performance of the prototype, we measure the temperature within the prototype as well as the driving time, and evaluate RSSI (Received Signal Strength Indicator) of wireless module in water. Then, we measure the swimming speed and swimming number to evaluate propulsion performance. As a result, developing the underwater robot with built-in driving system, driving in water, and wireless controlling by external computer have been realized.
In order to evaluate the intellectual productivity quantitatively, most conventional studies have utilized task performance of cognitive tasks. Meanwhile, more and more studies use physiological indices which reflect cognitive load so as to evaluate the intellectual productivity quantitatively. In this study, the method which evaluates task performance of intellectual workers by using several physiological indices (pupil diameter and heart rate variability) has been proposed. As estimation models of task performance, two machine learning models, Support Vector Regression (SVR) and Random Forests (RF), have been employed. As the result of a subject experiment, it was found that coefficient of determination (R2) of SVR was 0.875 and higher than that of RF (p<0.01). The result suggested that pupil diameter and heart rate variability were effective as the explanatory variables and SVR was also effective in task performance estimation.
This paper proposes an approach for automatic controller tuning based on experimental data only. It determines the controller parameters for a given cost function without knowing the true plant dynamics via Bayesian optimization. The validity of the proposed method is tested through some numerical simulations. Then, the proposed method is applied to PID controller tuning for an actual motor positioning system. The experimental results show that the proposed method improves the control performance effectively.
For the purpose of expanding the market of service robots, we developed a robot that realizes photography service by using common middleware and component system. For the task of photography, the system was constructed by grasping the functions to realize as modules, reusing the existing components, and developing the missing components. In order to improve reliability, we experimentally showed that it is possible to use two approaches, one that improves component performance and one that uses a combination of components. We also confirmed the effectiveness of the retry operation that human intervention to improve the certainty. In addition, for the purpose of commonality, we designed the specifications of components to cope with the differences in robot hardware and confirmed by experiment. As a result of the above, the success rate of the photography service can be improved to about 90% or more even in the case of two people in various environments.
This paper presents a continuous-time version of an optimization algorithm called Alternating Direction Method of Multipliers (ADMM), and analyzes convergence of the optimization dynamics based on passivity. First, a convex optimization problem is formulated as an equivalent ADMM form. We then present a novel continuous-time ADMM and prove convergence to a subset of optimal solutions of the convex optimization problem based on the theory of interconnected passive systems, where the cost function is assumed to be not strictly convex but just convex. Finally, the effectiveness of the present algorithm is demonstrated in a numerical simulation.
3D Measurement is an essential element of robotics for industrial usage. Many methods have been proposed so far, but some objects with specular reflections or subsurface scatterings are difficult to measure the shape accurately. We have developed a new method by using projector-camera Light Transport (LT) Matrix. LT Matrix is often used in computer graphics to make relighting images, light path understanding, and 3D measurement. It is useful for describing light paths on projector-camera system because it includes all light reflection information. However, it takes a long time to measure LT Matrix accurately. This paper first proposes how to obtain 3D shape from LT Matrix. Then we propose fast LT Matrix estimation method by using its sparseness. Our method can estimate accurate LT Matrix to obtain 3D shape in short time. Finally, we compare our results and previous 3D measurement methods, and we conclude our methods performs much better than other methods for metallic or transparent objects.
This paper is concerned with nonlinear analysis of a 1-d.o.f. vertical hopping robot, composed of its body, foot and a DC motor with crank mechanism. We show that its hopping motion under a constant voltage converges into a stable limit cycle, through physical experiments and numerical simulations. We then clarify this stabilization mechanism based on a simplified mathematical model, by showing that the negative torque-velocity correlation (weakness) of DC motors plays as a feedback law for stabilization. We also show that the limit cycles exhibit period-doubling bifurcation as the applied voltage increases, and the corresponding bifurcation diagram is affected by the weakness parameter of the DC motor.
In this paper, we model the delivery planning problem in a company conducting shipping business as a vehicle routing problem. According to an interview with this company, we formulate the problem based on the vehicle routing problem with time windows (VRPTW) with addition of lunch break constraints and multi-person work constraints. The quality of a routing plan is evaluated by the standard evaluation indices such as the number of vehicles and total operating time of the drivers with an optional evaluation index that evaluates the evenness of the operating time between the drivers. We also formulate a dynamic scheduling model, which is essential in practical delivery planning systems, to deal with additional orders that occur during delivery. We create a problem instance for the modeled problem that imitates the real situation in the company using actual map data, where the traveling time and distance between any two points is acquired from the digital map provided by ZENRIN co., ltd. In order to solve the problem instance we created, we propose a solution method based on both tabu search and iterated local search. In the experiment, we optimized this problem instance using the proposed method and evaluate the validity of the modeling.
To check whether a designed controller yields the desired response or not, we have to implement the controller and to execute the closed loop experiment. If we have an accurate model that reflects the actual dynamics of the plant, it is possible to see what happens in the closed loop before the designed controller is implemented. However, there are many cases in which it is difficult to obtain such a model due to some practical reasons. For example, it takes much time and cost for an ideal experiment to obtain a suitable model. In addition, there are some cases in which it is difficult to construct a model due to the complicated dynamics. For these cases, it is meaningful to develop a method for the prediction of the responses without a mathematical model of the plant. From these backgrounds, we present a new method for the prediction of the responses in a closed loop system by using only the experiment data and the controller which is to be implemented. The proposed method here can be regarded as a simple and innovative tool for a new framework of verifying the responses in the closed loop system without using the mathematical model of the plant. To show the validity of the proposed method, we also provide experimental results.