[in Japanese]

View full abstract

[in Japanese]

View full abstract

In recent years, a variety of human symbiotic robots has been developed. In many studies of the past, they have attempted to understand the speech and emotions of people. In this study, we focus on the interaction design that people understand robot mind and reply to the robot. For the design, we consider the psychological interaction utilizing baby talk. Here, when a robot has an appearance that can not be assumed to speak with baby talk, it is likely to feel uncomfortable by the negative adaptation gap. In this paper, we use a robot that has the appearance like a baby, to investigate the acceptable range of baby talk. Concretely, we report the result of the examination in which we created 49 videos of the robot having baby-like appearance using voice under 4 years old, and presented them in random order.

View full abstract

With the growth of robot technology, more educational-support robots, which support learning, are paid attention to. However, there is a problem in these robots. The problem is that a user loses his/her interest in the robot. To solve this problem, some studies which focus on a model of emotional expressions have been reported in Human-Agent-Interaction. A model of emotional expressions is defined as the agent expressing his/her emotion as with autonomous emotions. These models were shown to be beneficial to effective interaction between an agent and a human. Thus, this paper examines how psychological effect of robot which expresses the emotion by using the model of emotional expressions can prompt learners in a collaborative learning.

View full abstract

We propose a crossover operation for multi-terminal decision diagrams (MTBDDs). To survey this crossover operation, we conducted experiments of the evolution of MTBDDs with and without the proposed crossover operation. In an experiment on obtaining motion representation for a humanoid robot standing up, we confirmed that MTBDDs that have better fitness were obtained in the evolution of MTBDDs with the proposed crossover operation than without it. We also confirmed that MTBDDs that possess the smaller number of vertices were obtained in the evolution of MTBDDs with the proposed crossover operation than without it.

View full abstract

Two-wheeled inverted pendulum mobile robots have been developed so far as a transportation assistant system. The robot recognizes the user intention by the user's physical operation of the robot. It changes the control parameters according to the user intention. The conventional system fixes the number of user intentions beforehand, however, the number of intentions can vary according to a user preference. This paper proposes a data clustering system for user intention recognition and an incremental learning system of the intention recognition modules for the inverted mobile robot that is interactively operated by a user. It shows some experimental results with a real robot in order to show the validity of theproposed method.

View full abstract

Autonomous observation systems using a balloon or an airplane have been studied as a solution of information gathering systems from the sky. The balloon needs the helium gas reservation and a relatively long time and specialists of gas maintenance for the ight preparations. Furthermore, the balloon system is not available under strong wind condition. On the other hand, an airplane system needs less time for ight preparations, but long-term activity is difficult due to limitation of fuel. In order to complement those information gathering system, we have proposed a tethered ying robot based on a kite that ies with wind power as one of the natural power sources and conducted some experiments with a real robot we designed and built. It is difficult to design the ight controller because wind situation often varies in real robot experimentation. We have also developed a computational dynamics simulator for the kite-based tethered ying robot.
 In the paper [1], authors designed 3-inputs 1-output fuzzy controller in order to reect control strategy based on human operation. In real robot experimentation, we conducted ight controller based on fuzzy control theory using fuzzy rule table which is written by human. However, there is a possibility that this fuzzy table differs from the actual human operation. Furthermore, it becomes difficult to define the table by manpower if the number of state variables or membership functions becomes big. We verify a learning method of fuzzy control parameters for the robot using human operation data. The original controller for the kite-based tethered ying robot was designed based on 3-inputs 1-output fuzzy controller in order to reect control strategy based on human operation. However, as the controller cannot have control for extending the tether line to control the kite stably in condition of various change of wind because of lack of information for state description.
 This paper aims at acquiring human control strategy for the kite-based ying robot using human operation data. Besides learning the fuzzy control parameters, we also use k-nearest neighbor algorithm and artificial neural network. We extend the fuzzy controller with 4-inputs 1-output system in order to reect human control strategy and verify their effectiveness with the computational simulations. The weighted k-nearest neighbor algorithm that showed the best performance has a disadvantage of heavy calculation processing if training data is big, because this technique explores all of the training data. We propose a method that reduces training data itself and also shows its validity with the computational simulations.

View full abstract

In recent years, robots have served conspicuously in dangerous environments including disaster areas and outer space. Within such environments, however, robots suddenly may fall into dangerous situations when commands from people to avoid certain risks don't reach them in time. Accordingly acquisition of autonomous risk-avoidance behavior in robots is required. It is thought that the technique for realizing this capability will make use of reinforcement learning. As a new reinforcement-learning framework for avoiding risk, we propose a probability-based reinforcement learning method and apply it to behavior acquisition in robots.

View full abstract

Myoelectric prosthetic hands are desired by those who have lost their hands not only for decoration but also for the functions to help them with their activities of daily living (ADL). The prosthetic robotic hands for daily use should be little stressful to the user by weight reduction, and be fault-tolerant. It is thus required to simplify the prosthetic robotic hand while maintaining its grasp function. The aim of this study is therefore to develop a simplified prosthetic hand to realize the most important hand motions in ADL by trading off the cost and performance. This paper reports the development of a two degree-of-freedom prosthetic hand with two motors to drive and the carpometacarpal joint of the thumb and the interlocked meta phalangeal joints of the other four fingers. In order to optimize the structure of the fingers, the model of the prosthetic hand was proposed and 7 hands with different structural parameters were manufactured and tested in a pick-and-place experiment with objects of daily use. The proper parameters suggest by correspondence analysis were adopted to design the prosthetic hand for a subject with upper limb amputation. The pick-and-place experiment by the subject showed the applicability of the structure and the effect of weight reduction.

View full abstract

People experience feelings of indebtedness when receiving help from an agent. This feeling of indebtedness has a potentially negative impact on the human attitude. However, few experimental analyses this evaluating effect of the help an agent is available. The aim of this study is to design an agent that can assist people without making them feel indebted to that agent. The present research proposed a help method where the agent helps people to achieve their goal using a different approach from the recipient and verified its effectiveness by experiment. It was established that an agent can assist recipients without adverse effects if the help method is different.

View full abstract

This paper describes system improvement of P-CUBE and a study conducted on the usefulness of P-CUBE as an education tool for a programming beginner. P-CUBE is a block type programming tool for beginners including visually impaired. We report the improvement of the programming blocks. Then, we conducted an experiment for comparing P-CUBE and conventional programming software for a mobile robot. We recorded the times of referencing the manual book and the time required for programming exercise during the experiment. The result showed that P-CUBE is useful as a programming education tool for beginners.

View full abstract

Emotion is one of the important roles in communication. Emotion recognition by the computer has been studied. In this study, we tried to recognize the emotion that represents a strong emotional behavior and physiological changes in a short period of time. The body's movements as non-verbal behavior were measured by wearing a pressure-sensitive conductive clothes same as a feeling of common clothes. The emotion was recorded by the circumplex model of Russell represented on two axes of pleasure-unpleasure and arousing-sleepy. Emotion was recognized using machine learning by recording the movement and the emotion.

View full abstract

Data mining has emerged as a significant technology for discovering knowledge in large quantities of data. When we mine databases including private information, it is necessary to process data mining in a privacy-preserving manner, and privacy-preserving data mining has received a growing amount of attention in recent years. In this paper, we propose a method to extract global fuzzy rules from distributed data with the same attributes in a privacy-preserving manner. This method transfers only values necessary for the extraction process without collecting any data at one place and can obtain the global fuzzy rules at all places where distributed data exist. Each distributed data set can be characterized by comparing the local fuzzy rules to the global ones. We illustrate a result using Wine Data from UCI Machine Learning Repository.

View full abstract

In this study, we construct a system which recognizes and visualizes batting swing patterns in baseball using a motion sensor. In fact, in baseball practice, swing differences occur between the swing patterns. We propose a method to evaluate the swing differences quantitatively by constructing visualization system. In the proposed system, support vector machine (SVM) is valid method for pattern recognition of batting swings. Furthermore, visualization by principal component analysis (PCA) enables us to evaluate the swing differences quantitatively. Through the experiments, we confirm that the visualization system is available as a technical support tool for the batters.

View full abstract

Fuzzy Q-learning has been studied because it can treat a continuous state that can not be done by Q-learning. Fuzzy Q-learning by Glorennec is used for many related studies. We propose a method that improves Glorennec's Fuzzy Q-learning. We apply the method to the pursuit problem in a continuous environment and compare with the Glorennec's one.

View full abstract