Recently, robots with many degrees of freedom to achieve various tasks have been developed, and learning based control for these robots have attracted considerable attention. However, for these robots, there are two significant problems: real-time learning and a lack of generalization ability. In this paper, to solve these problems, we propose to design a body for abstracting the necessary small state-action space from a huge state-action space by utilizing properties of the real world, like dynamics, mechanical constraints, and so on. As an effective example, we consider grasping and carrying tasks and develop a redundant manipulator inspired by an octopus. We show that by designing the manipulator to utilize properties of the real world, the state-action space can be abstracted, and the real-time learning and lack of generalization ability problems can be solved.
Nowadays mobile phones with some motion sensors like an accelerometer or a gyroscope are becoming widespread. Under this situation, there are various studies to identify humans or their activities through the data measured by the motion sensors. In this study we focus on identifying humans using data from an accelerometer. In many current studies that deal with human identification using an accelerometer, it is assumed that humans walk only on the flat floor. Here, we propose a two stage technique to identify not only humans but also their three walkingstates (‘walking on the flats’, ‘going up stairs’ and ‘going down stairs’). Specifically our method identifies the walking state at the first stage and subsequently identify humans based on the specific classifier for the respective walkingstates. We employed five classifiers (k-Nearest Neighbor, Classification And Regressive Trees, Na ve Bayes Classifier, Linear Discriminant Analysis and Support Vector Machine). From the results of the experiments with data from 10 human subjects, the best walking state identification of 95.7% accuracy was achieved by Linear Discriminant Analysis. Also, in case of human identification, the best performances obtained are: 85.0% accuracy for ‘walking on the flats’, 90.0% accuracy for ‘going up stairs’ and 77.0% accuracy ‘going down stairs’ with Linear Discriminant Analysis. The best human identification result with two stage process was obtained as 80.4%. It has also been found that for all the classifiers except k-Nearest Neighbor, walking states affect the identification result of a specific classifier and the two stage process (identifying walking state before identifying the human) of identification produce better result in terms of identification accuracy.
Elementary, junior high or high school students are required to understand ways of thinking against questions that involve calculation in the process of solution. Students who know formula only have to remember all formula to solve various questions and cannot have ability to apply one to similar questions. In this study, a learning system that assists establishing ways of thinking against percentage and speed questions by intuitive understanding iterations is proposed. According to experimental results, the proposed system that puts emphasis on ways of thinking was effective to learning and its radication rather than the comparative system that puts emphasis on solution by formula.
Ordinary learning methods need to design indexes by a human being to learn a given task; teaching signals, a fitness function or a reward function. Generally they are designed for each task and environment so that they must be re-designed if tasks or environment change. Our aim is to design universal indexes for learning of a given task, which is not needed to re-design for different tasks or environment. To realize it, we propose the evaluations for each sensory information based on organism's pleasure and pain, and the method to generate a reward by the evaluations. By refering for organism's response, the propsed method can generate a proper reward with the intuitive interaction with a human being. Our experiment confirmed that plural participants could interact with a robot and make the robot achieving a task.
In this paper, we propose a method of data mining in education to solve the following problems. As one of the problems in university, there is tendency to increase the dropout students, since many high school students did not have clear reason to enroll in university. To decrease them, in the high school, teachers have to change their career perception and to increase their motivation to enroll in university. To perform these educations effectively, it is desirable for teachers to make the instructional design, after teachers appropriately grasp the change of student's career perception depending on the student's characters. Therefore, we propose a method to solve the above issue by using Support Vector Machine (SVM) and calculating gradient. Specifically, we use the independent variables as career perceptions and dependent variables as motivation to enroll in university to construct SVM. Moreover, by calculating gradient of a function for motivation provided by SVM, we get direction vector to increase their motivation. By performing instructional design following the obtained direction vector, it is desirable to increase the student's motivation to enroll in university. Moreover, we describe the prototype of the system to which we applied our proposed method.
Robots require environmental maps to move autonomously in the human living environment. Techniques for constructing environmental maps which have been proposed in the past mostly construct environmental maps based on shape information alone. However, when only shape information is used there is the problem that, in uniform environments which have few geometrical features, such as corridors, multiple candidates for the robot's own position are generated and its estimation is difficult. The present paper therefore focused on visual information as useful data to estimate the robot's own position, and investigated a self-location estimation method using only image information. For the proposed technique, learning was conducted using the SOM algorithm for unsupervised learning of omnidirectional image data, obtained in the environment in advance, and a self-location discriminator was built. The robot determines, by means of the location estimation discriminator, in which area in the environment the images it observed while moving could have been taken. In this article we have verified the usability of the location estimation discriminator for an indoor environment with many geometrical features and rich in visual information such as patterns and textures, and for a corridor environment with few geometrical features and poor in visual information.
Data analysis with text mining is now attracting many people. Many softwares of text mining are also available. Users of text mining softwares need to acquire text mining skills to select and operate tools, to analyze data, and so on. In order to acquire the skills, users have to be assigned exercises about selection and operation of suitable tools, and interpretation of analyzed data. So, this paper proposes a tutorial system for users to acquire the text mining skills with a software TETDM. The tutorial system gives users exercises. The exercises are ordered according to its level: basic one is arranged in the former and advanced one is in the latter. We evaluated the proposed tutorial system in an evaluation experiment. It was confirmed that the subjects of the experiment could have acquired text mining skills with TETDM by the tutorial.
Tingyi (Cayman Islands) Holding Corp entered the Mainland China market in 1992. In a short period of time, the company had good performance in instant noodles and beverage departments and gained the top position in the China market. One of the success factors was that Tingyi entered the China market earlier than other competitors, so got a perfect position, and built its brand image. On the other hand, Tingyi took a chance by grasping the opportunity presented by the development of the fast-food industry and supermarkets. Tingyi has started business partnerships with Japanese food companies continuously since 1996, These include Sanyo Foods Co., Ltd, Asahi Breweries, Ltd, Itochu Corporation, Pasco Shikishima Corporation, Nippon Flour Mills Co., Ltd, Calbee, Inc, etc. Tingyi also made partnerships with PepsiCo Inc of China to expand its business range.According to statistics, Tingyi reached a 47% share in China's instant noodle market in 2014, close to half of the entire China market. Profitability is a key which determines the value of a corporation, and when the market slumps, it can show that an outstanding enterprise has more competitive advantage than others. In this paper, we try to use the fuzzy VRIO, FIVE FORCES and SWOT methods to analyze the internal and external factors of Tingyi, find threats and opportunities that the enterprise now faces, and identify how Tingyi can keep its competitive advantage in China. In this paper, in order to analyze the current management strategy of Tingyi, we use the Five Forces method to analyze the external environment, and the Fuzzy VRIO method on the internal environment. The strengths and weaknesses of Tingyi's internal environment, and the opportunities and threats from the external are revealed using a SWOT analysis. In the future, we believe it is essential for Tingyi to formulate effective strategies to continue their dominance of the Chinese market.
In Japan, many of the highway bridges intensively developed during the high economic growth period are deteriorating; it is the urgent need to develop a proper repair plan to extend the service life of those bridges. To develop proper inspection and repair plans and to select a proper construction method, it is necessary to precisely grasp the geometry of the current status. However, in performing the resurvey work on the site, accompanying actions such as suspension of traffic become necessary. Therefore, it is difficult to conduct the work over about 700,000 bridges across the nation in terms of costs and human resources. To solve these problems, a method of visualizing the current status for the superstructure of the highway bridge three-dimensionally using MMS, which does not require suspension of traffic, has been proposed. However, as MMS can measure only the area facing the road, the generated geometry is a surface model represented with polylines without thickness, which is not designed to represent the whole superstructure of bridge. In addition, it is impossible to make the product model of a bridge substructure. Thus in this research, we propose a method of generating a three-dimensional (3D) whole model of the superstructure and substructure of a bridge from the point cloud data measured using the ground-installationtype laser scanner and UAV.
A human detection method for omnidirectional images with distortion and appearance change is proposed in order to improve detection accuracy without camera calibration. The method is based on deep convolutional neural network, and a training data increase method is also proposed with applying translation, scaling, rotation, and brightness change operators to manually prepared training data. In the evaluation experiment with actual omnidirectional images, under the condition at false positive per detection window 0.001, the proposal achieves miss rate 28.2% and miss rate of a combination of HOG and Real AdaBoost, which is standard method in object detection, is 77.5%. In addition, it is validated that the proposed training data increase method improves detection accuracy, and relationships between visualized feature maps and detection accuracy are discussed. The proposal provides improvement of human detection accuracy without estimating camera parameters, where camera calibration is difficult in situations such as large number of cameras, cameras on distant places, and unknown camera models.