Search

We have already confirmed that the artificial subtle expressions (ASEs) from a robot can accurately and intuitively convey its internal states to participants [10]. In this paper, we then experimentally investigated whether the ASEs from an on-screen artifact could also convey the artifact's internal states to participants in order to confirm whether the ASEs can be consistently interpreted regardless of the types of artifacts. The results clearly showed that the ASEs expressed from an on-screen artifact succeeded in accurately and intuitively conveying the artifact's internal states to the participants. Therefore, we confirmed that the ASEs' interpretations were consistent regardless of the types of artifacts.

In our work on evolutionary robotics, two different approaches have been developed: simulation-based behavior learning with genetic algorithm and direct teaching with a learning classifier system in real environments. As the first approach, action-based environment recognition for an autonomous mobile robot. The robot moves in various environments by executing behaviors designed by a human designer and obtains different action sequences. These action sequences are automatically classified by self-organizing maps and the structure of the environment is identified from them. We also developed a GA-based behavior learning method in which a robot can learn suitable behaviors to recognize, and conducted simulation experiments to verify the learning ability. However, all the experiments has been done through only computer simulation. Thus we attempted to develop a direct teaching framework in which a real robot learned from human teacher using LCS in real and physical environments. Direct teaching means a human teacher gives adequate actions to a mobile robot by a GUI at work, and this teaching can accelerate a robot to learn classifiers. This framework is important to realize evolutionary robotics which can learn sufficiently fast in real environments, and we confirmed that it is useful by experimenting with a small mobile robot. In this chapter, we describe these two innovative approaches in evolutionary robotics in detail and discuss them in various points of view.

This paper describes a novel remote control operable with stroking its surface. Advantages of the developed remote control are high familiarity and stroke operation. Those enable users to have familiarity with it and to use it without looking at the fingers. We apply it to an interaction system with TV. The proposed system has the tolerance for mistakes in comparison with conventional button-based remote controls because it enables unfamiliar users to home electric appliances to use it casually without fear of mistakes and unexpected behavior.

We propose a framework to extract topic maps from a set of Web pages. We use the clustering method with the Web pages and extract the topic map prototypes. We introduced the following two points to the existing clustering method: The first is merging only the linked Web pages, thus extracting the underlying relationships between the topics. The second is introducing weighting based on similarity from the contents of the Web pages and relevance between topics of pages. The relevance is based on the types of links with directories in Web sites structure and the distance between the directories in which the pages are located. We generate the topic map prototypes from the results of the clustering. Finally, users complete the prototype by labeling the topics and associations and removing the unnecessary items. For this paper, at the first step, we mounted the proposed clustering method and extracted the prototype with the method.

In this paper, we present a human-robot teaching framework that uses "virtual" games as a means for adapting a robot to its user through natural interaction in a controlled environment. We present an experimental study in which participants instruct an AIBO pet robot while playing different games together on a computer generated playfield. By playing the games in cooperation with its user, the robot learns to understand the user's natural way of giving multimodal positive and negative feedback. The games are designed in a way that the robot can reliably anticipate positive or negative feedback based on the game state and freely explore its user's reward behavior by making good or bad moves. We implemented a two-staged learning method combining Hidden Markov Models and a mathematical model of classical conditioning to learn how to discriminate between positive and negative feedback. After finishing the training the system was able to recognize positive and negative reward based on speech and touch with an average accuracy of 90.33%.

This paper reports a new document retrieval method using non-relevant documents. Suppose, we need to find documents interesting to the user in as few iterations of human intervention as possible. In each iteration, a relatively small set of documents is evaluated in terms of the relevance to the user's interest. Ordinary relevance feedback needs both relevant and non-relevant documents, but the initial set of documents checked by the user may often not include relevant documents. Accordingly we propose a new feedback method using non-relevant documents only. This "non-relevance feedback" selects documents classified as "not non-relevant" and close to the boundary defined by the discriminant function obtained from one-class SVM. Experiments show that this method can efficiently retrieve a relevant documents.

This paper proposes a novel interaction model of a human-robot cooperative task. The model employs CEA (Commands Embedded in Actions), which reduces a human cognitive load because it requires less explicit human-robot communication than direct commanding methods in conventional interaction models. We propose a guideline along which to design robots' actions based on CEA, and apply it to a cooperative sweeping task by a human and a small mobile robot. CEA is experimentally shown to reduce the human cognitive load more than direct commanding methods do in this sweeping task.

This paper reports a new document retrieval method using non-relevant documents. From a large data set of documents, we need to find documents that relate to human interesting in as few iterations of human testing or checking as possible. In each iteration a comparatively small batch of documents is evaluated for relating to the human interesting. The relevance feedback needs a set of relevant and non-relevant documents to work usefully. However, the initial retrieved documents, which are displayed to a user, sometimes don't include relevant documents. In order to solve this problem, we propose a new feedback method using information of non-relevant documents only. We named this method non-relevance feedback document retrieval. The non-relevance feedback document retrieval is based on One-class Support Vector Machine. Our experimental results show that this method can retrieve relevant documents using information of non-relevant documents only.

With the dramatic development of web technologies, tremendous amount of information become available to users. The great advantages of the web are the ease with which information can be published and made available to a wide audience, and the ability to organize and connect different resources in a graph-based structure using hyperlinks. However, most of these links are created manually and the page that the link represents must be known to the author of the link. In this paper, we propose a decision-tree-based approach to solve this problem. We set up a system that gathers information about the candidate pages, evaluates them and creates links to them automatically.

Query expansion techniques generally select new query terms from a set of top ranked documents. Although a user’s manual judgment of those documents would much help to select good expansion terms, it is difficult to get enough feedback from users in practical situations. In this paper we propose a query expansion technique which performs well even if a user notifies just a relevant document and a non-relevant document. In order to tackle this specific condition, we introduce two refinements to a well-known query expansion technique. One is to increase documents possibly being relevant by a transductive learning method because the more relevant documents will produce the better performance. The other is a modified term scoring scheme based on the results of the learning method and a simple function. Experimental results show that our technique outperforms some traditional methods in standard precision and recall criteria.

This paper describes a human-agent interaction in which a user and a life-like agent mutually acquire the other's mind mapping through a mutual mind reading game. In these several years, a lot of studies have been done on a life-like agent such a Micro Soft agent, an interface agent. Through development of various life-like agents, a mind like emotion, processing load has been recognized to play an important role in making them believable to a user. For establishing effective and natural communication between a agent and a user, they need to read the other's mind from expressions and we call the mapping from expressions to mind states mind mapping. If an agent and a user don't obtain these mind mappings, they can not utilize behaviors which significantly depend on the other's mind.We formalize such mutual mind reading and propose a framework in which a user and a life-like agent mutually acquire mind mappings each other. In our framework, a user plays a mutual mind reading game with an agent and they gradually learn to read the other's mind through the game. Eventually we fully implement our framework and make experiments to investigate its effectiveness.

This paper describes an automatic monitoring system that constantly checks partial updates in Web pages and notifies the user about them. While one of the most important advantages of the WWW is frequent updates of Web pages, we need to constantly check them out and this task may take much cognitive load. Unfortunately applications to automatically check such updates cannot deal with partial updates like updates in a particular cell of a table in a Web page. Hence we developed an automatic monitoring system that checks such partial updates. The user can give a systemregions in which he/she wants to know the updates in a Web page as training examples, and the system is able to learn rules to identify the partial updates by relational learning. We implemented the system and some executed examples were presented.

This paper describes a system for collecting Web pages that are relevant to a particular topic through an interactive approach. Indicated some relevant pages by a user, this system automatically constructs a set of rules to find new relevant pages. The purpose of the system is to reduce users' browsing cost by filtering non-relevant pages automatically. Such an approach can be useful when users do not know how to describe their requirements to search engines. We describe the representation and the learning algorithm, and also show the experiments comparing its performance with a search engine.

This paper describes an evolutionary way to lean behaviors of a mobile robot for recognizing environments. We have proposed AEM (Action-based Environment Modeling) which is an appropriate approach for a simple mobile robot to recognize environments, and made experiments using a real robot. The suitable behaviors for AEM have been described by a human designer. However the design is very difficult for them because of the huge search space. Thus we propose the evolutionary design method of such behaviors using genetic algorithm and make experiments in which a robot recognizes the environments with different structures. As results, we found out that the evolutionary approach is promising to automatically acquire behaviors for AEM.