We develop a novel jet engine health monitoring method using a high-fidelity, physics-based simulation model. In jet engine health monitoring, fault detection with a state estimation is frequently carried out. However, a challenge for a filtering-based health monitoring method with a high-fidelity model is the difficulty to analytically obtain the Jacobians, which is required for the extended Kalman filter. Our approach is to derive a linearized model a priori at a fixed operation point, and use a robust filter method to guarantee a reliable health monitoring over a wide range of operation points despite the linearization errors. We validate the proposed method by extensive simulations.
Recently, the framework of shared control between a human operator and a robot system using a cooperative controller with human operator to decide control input to a robot system has been drawing much attention for dexterous telemanipulation. However, desiging a suitable controller for shared control scheme for the given task is difficult due to the difficulties in modeling of operator's behavior and environment. In this paper,we proposed a model-free approach using reinforcement learning to learn a shared control policy through interactions with the operator, robot and environment. To validate our method,we adopted a page turning task by telemanipulation and developed an experimental platform with a physical simulator.Experimental results suggest that our method is able to learn task-relevant shared control for flexible and enhanced dexterous manipulation.
Our previous study designed interactive floor projection system with hands and toes input in terms of quick and light input system. This system lets user interact with floor projection by pointing the finger or stepping on projection image. It allows to realize floor interaction without using even a single hands.
However, in this system, it is difficult for user to input by toes to desired position precisely because of low accuracy of toe detection. In this paper, we propose a new toe detection method by means of implementing a leg model to improve the accuracy of toe detection. The model signifies human leg. It consists of three joints and; thigh, cnemis and foot length. Moreover, we conduct user study to investigate accuracy of toe detection compared with proposed and previous method. As a result, proposed method can detect toe position with higher accuracy than previous method.
A new algorithm is proposed to estimate parameters of MIMO (Multiple-Input Multiple-Output) system. The algorithm is an expansion of the previously proposed algorithm by the author for SISO (Single-Input Single-Output) system and MISO (Multiple-Input Single-Output) system. Those algorithms use an iterative calculation and they consist of two parts. One part is the estimation of the system parameters using the variances of errors. The other part is the estimation of the variances of errors using the system parameters. The estimation of the system parameters is one of the least squares identification algorithms using eigenvector. The variances of errors are estimated by solving linear simultaneous equations derived from the transfer functions of the system. Some simulations show the effectiveness of the proposed method.
The lack of helpers in facilities to take care of the elderly has become a serious problem in Japan,because of the rapid increase of the elderly. It is necessary for the elderly living alone to keep their living willingness enough to stay in a status free from case. The method uses the brightness distribution sensors to collect movement logs in order to protect the privacy of the elderly. The brightness distribution acquired by the sensors brings information significant enough for a machine to discern living activities. This study figures out the conscientious degree with the difference of the body trunk movement of the person carrying out of the living activities. In an experiment for the elderly, the f-measure with which the method has recognized activities of cleaning, cooking,and washing are 0.975, 0.912, and 0.927, respectively. The experiment shows 0.599, which indicates conscientious degree of cleaning. It justifies the proposed method uses the movement of body trunk to figuring out the conscientious degree.
In recent years, cooperative control of multi-agents, especially formation control, has attracted considerable interest among control engineers. As one of the formation control methods, we deal with the leader-following formation navigation (LFFN). The LFFN is defined such that a human operator or an autonomous leader agent forms a queue of agents to guide multiple followers. The followers pass along the leader's trajectory with the same motion as the leader. When the leader passes along a safe route to avoid obstacles, the followers also pass along a similar safe trajectory. However, when using the LFFN, the followers can shape only a single column formation.
In this study, we first extend the above LFFN's usage to parallel single columns using adjoining virtual leaders. Next, we consider virtual leaders adjoining the leader to design virtual trajectories along which virtual leaders passed and virtual leader's velocities on their path. We design the followers' target points which track the virtual leaders path with designed virtual velocity. Using our proposed method, the followers shape a formation which is not a single column but a multiple column type. Finally, we demonstrate the effectiveness of our newly proposed method through simulation and experimentation.