In this paper, we propose a novel method for a robot to detect robot-directed speech: to distinguish speech that users speak to a robot from speech that users speak to other people or to themselves. The originality of this work is the introduction of a multimodal semantic confidence (MSC) measure, which is used for domain classification of input speech based on the decision on whether the speech can be interpreted as a feasible action under the current physical situation in an object manipulation task. This measure is calculated by integrating speech, object, and motion confidence with weightings that are optimized by logistic regression. Then we integrate this measure with gaze tracking and conduct experiments under conditions of natural human-robot interactions. Experimental results show that the proposed method achieves a high performance of 94% and 96% in average recall and precision rates, respectively, for robot-directed speech detection.

抄録全体を表示

In this paper, we proposed the user uninterruptibility estimation method based on focused Application-Switching (AS) during PC work for establishing information display timing control scheme with less intelligent activity disturbance for users. At first, we collected and analyzed the PC operation records and the subjective uninterruptibility of users. From the analysis, we selected features in AS timing that affect user uninterruptibility. Then, we provided the estimation method based on co-occurring features that are observed in AS timing, and confirmed the availability of our method.

抄録全体を表示

This paper describes a novel methodology, namely ``Coaching'', which allows humans to give a subjective evaluation to an agent in an iterative manner. This is an interactive learning method to improve the reinforcement learning by modifying a reward function dynamically according to given evaluations by a trainer and the learning situation of the agent. We demonstrate that the agent can learn different reward functions by given instructions such as ``good or bad'' by human's observation, and can also obtain a set of behavior based on the learnt reward functions through several experiments.

抄録全体を表示

In the future, robots will support our work in our daily life. We believe that robots should learn desirable behavior through human-robot interaction. However, it is hard for humans to instruct the robots on all actions. It therefore is important that the robots can utilize rewards (evaluations) as well as instructions to reduce humans' efforts. Additionally, ``intervals'' which allow humans to give instructions and evaluations are also important because there are delays in giving them. We hence focused on ``delays in initiating actions of a robot'' and proposed a method of changing them according to the progress of learning: long delays at early stages, and short at later stages. In other words, if a robot is not sure about its action, it initiates the action laggardly, but if it is confident about its action, it initiates the action immediately. In this work, we conducted experiments on teaching AIBO to shake hands using instructions and evaluations under two conditions: Varying Condition under which the delays vary in accordance with the progress of learning, and Constant Condition under which the delays are set at medium constant. The result demonstrated that Varying Condition improves learning efficiency significantly and impresses humans as teachable.

抄録全体を表示

The goal of this paper is to link a bridge between social relationship and cultural variation to predict conversants' non-verbal behaviors. This idea serves as a basis of establishing a parameter based socio-cultural model, which determines non-verbal expressive parameters that specify the shapes of agent's nonverbal behaviors in HAI. As the first step, a comparative corpus analysis is done for two cultures in two specific social relationships. Next, by integrating the cultural and social parameters factors with the empirical data from corpus analysis, we establish a model that predicts posture. The predictions from our model successfully demonstrate that both cultural background and social relationship moderate communicative non-verbal behaviors.

抄録全体を表示

In this study, we aim at clarification of the factor that promotes an user's acceptance of suggestion from an interactive agent in driving environment. Our aim is to figure out how human beings accept the encouragement from interaction objects, and also which kinds of dialogues or action controls are necessary for the design of car navigation system which makes suggestion and requests to drivers. Firstly, we had an experiment for collecting dialogue between humans in driving simulation environment, then we analyzed the drivers' acceptance and evaluation for the navigators. As the results, we found that the presence and reliability of the navigator highly relate to the acceptance of suggestion from the navigator. When navigators were next to drivers, the rate of drivers' suggestion acceptance rose. However, the stress of drivers increased. In addition, based on the linguistic and acoustic analysis of the navigators' utterances, we found out some points of designing system utterance of suggestion to promote user's acceptance. We found that expressing the grounds of suggestions, showing the exact numbers, and the wide pitch ranges, all highly relate to the acceptance of suggestions.

抄録全体を表示

We describe artificial subtle expressions (ASEs) as intuitive notification methodology for artificial agents especially in order to convey their internal states for users. We prepared two types of audio ASEs; one is a flat artificial sound (flat ASE), and the other is a decreasing sound (decreasing ASE). These two ASEs were played after a robot made a suggestion to the users. Specifically, we expected that the decreasing ASE will inform users of the robot's lower confidence about the suggestions. We then conducted a simple experiment to observe whether the participants accepted or rejected the robot's suggestion in terms of the ASEs. The result showed that they accepted the robot's suggestion when the flat ASE was used, while they rejected it when the decreasing ASE was used. Therefore, we found that the ASEs succeeded in conveying the robot's internal state to the users accurately and intuitively.

抄録全体を表示

We propose an informative DOM node extraction method from a Web page for preprocessing of Web content mining. Our proposed method LM uses layout data of DOM nodes generated by a generic Web browser, and the learning set consists of hundreds of Web pages and the annotations of informative DOM nodes of those Web pages. Our method does not require large scale crawling of the whole Web site to which the target Web page belongs. We design LM so that it uses the information of the learning set more efficiently in comparison to the existing method that uses the same learning set. By experiments, we evaluate the methods obtained by combining one that consists of the method for extracting the informative DOM node both the proposed method and the existing methods, and the existing noise elimination methods: Heur removes advertisements and link-lists by some heuristics and CE removes the DOM nodes existing in the Web pages in the same Web site to which the target Web page belongs. Experimental results show that 1) LM outperforms other methods for extracting the informative DOM node, 2) the combination method (LM, {CE(10), Heur}) based on LM (precision: 0.755, recall: 0.826, F-measure: 0.746) outperforms other combination methods.

抄録全体を表示