The general idea about the human-robot conversations is that the utterances of a robot should be easy to understand. It has been considered robots should provide the complete information, without the necessity of a human asking further questions. On the other hand, considering the human-human conversation, in particular, between a child and a caregiver; even though the child has a limited speaking ability to tell a story entirely, child and caregiver can build deep bonds through the aggressive support of the caregivers. This study aims to analyze the human behaviors when a robot speaks to a human with insufficient words. We built a syntactic method to create a deficiency in the elements of the robot’s utterances. We analyzed the human replies in such a human-robot conversation. The results showed that when the robot used “Kotobatarazu Utterance, ” the human partners increased asking questions to the robot, talked more about the details of the topic and they decreased slight responses. These results indicate that a human can join more actively to the conversation when the robot provides incompleteness in its conversation. Furthermore, the recorded interaction between a human and the “Kotobatarazu” robot was evaluated as cooperative.

View full abstract

Recently, text chat has become the mainstream in CMC(Computer-Mediated Communication), and it is used even for discussion to build consensus. In the case of CMC between strangers, anonymity is attracting attention as factor influencing communication. However, influence of anonymity on CMC between acquaintances has not been clarified. Therefore, with the aim of developing a new text chat system that supports discussion between acquaintances, we examine the influence of anonymity in text chat on consensus building between acquaintances. As a result of the experiment, in case of between acquaintances, there is a possibility that the time to consensus building become shorter in discussion by anonymity than in discussion by real name. Moreover, there is a possibility that discussion by anonymity is more efficient than discussion by real name.

View full abstract

We have studied the screening tool for dementia patients using a P300-based spelling-BCI to develop a new cognitive function evaluation method. We quantified a low attention with dementia using the average distance from the incorrect characters to the correct characters on the Japanese alphabet screen. We called the character distance to Spelling-Error Distance Value (SEDV). We conducted experiments for 24 elderly subjects, and performed regression analysis between SEDV and neuropsychological testing (MMSE, FAB, TMT and DST). As the results, there was a significant negative correlation between MMSE and SEDV (r = −0.41, p < 0.05), and there was a significant negative correlation between FAB and SEDV (r = −0.65, p < 0.001). Next, 24 subjects divided into five groups according to the MMSE scores. There was a difference of SEDV 2.58 characters between MMSE average of 10.5 point group and MMSE average score of 27.6 point group, 2.89 characters between the 10.5 point group and the 24.9 point group. Therefore, SEDV could be an index of the low attention with dementia in the limited MMSE average group. In addition, because of significant negative correlation between FAB and SEDV in less than 24 point of MMSE (r = −0.95, p < 0.05), SEDV could be an index of frontal lobe function deterioration limitedly.

View full abstract

We often found in shopping malls that parents struggle to take care of their children. According to our survey, 85% of parents find it difficult to shop with their children in shopping malls. We think that the children’s lack of purpose makes them bored and do things that irritate their parents. We introduce a treasure hunting game to solve this problem. We conducted two field studies to design and test our game approach. Through the first study, we found that the game had success to give children a particular purpose, however, the game also can be an obstacle for shopping for parents. In response to this findings, we redesigned our game and conducted a second study to verify the new game design. The result of the second study revealed that the redesigned treasure hunting game allowed both parents and children to achieve their individual purposes.

View full abstract

In this paper, we propose a method of sitting posture classification with chair-shaped device, called SenseCaster, which non-intrusively measures a user's center of pressure and weight using a pressure sensor embedded inside caster. We also propose a method of estimating a seat rotation with neodymium magnet and magnetic sensors to calculate the center of pressure. We implemented and conducted several experiments to confirm SenseCaster’s functionality. In the first experiment, we evaluate an accuracy of measuring a user’s center of pressure and weight. The mean absolute error was (2.7, 3.3) mm and 0.31 kg, and repeatability error was (1.3, 1.8) mm and 0.10 kg. In the second experiment, we classified ten simple sitting postures using the center of pressure and the weight data of 14 subjects. The average accuracy was 0.80 for familiar users (SenseCaster had collected their data before) and 0.70 for unfamiliar users.

View full abstract

We propose BlowFab, a prototyping method used to create a 2.5- dimensional prototype in a short time by combining laser cutting and blow molding techniques. The user creates adhesive areas and inflatable areas by engraving and cutting multilayered plastic sheets using a laser cutter. These adhesive areas are fused automatically by overlapping two crafted sheets and softening them with a heater. The user can then create hard prototypes by injecting air into the sheets. Objects can be bent in any direction by cutting incisions or engraving a resistant resin. The user can create uneven textures by engraving a pattern with a heat-resistant lm. In this study, the design process is described using the proposed method. The study also evaluates possible bending mechanisms and texture expression methods along with various usage scenarios and discusses the resolution, strength, and reusability of the prototype developed.

View full abstract

Gesture-based recognition is one of the most intuitive methods for inputting information and is not subject to cumbersome operations. Recognition is performed on human’s consecutive motion without reference to retrial or alteration by user. We propose a gesture recognition model with a mechanism for correcting recognition errors that operates interactively and is practical. We applied the model to a setting involving a manual grading task in order to verify its effectiveness. Our system, named GERMIC, consists of two major modules, namely, handwritten recognition and interactive correction. Recognition is materialized with image feature extraction and convolutional neural network. A mechanism for interactive correction is called on-demand by a user-based trigger. GERMIC monitors, track, and stores information on the user’s grading task and generates output based on the recognition information collected. In contrast to conventional grading done manually, GERMIC significantly shortens the total time for completing the task by 24.7% and demonstrates the effectiveness of the model with interactive correction in two real world user environments.

View full abstract

In this study, we focused on domain-specific emotions and investigated the characteristics of the emotions evoked in the individual voluntary manufacture and the characteristics of the manufacture itself. In study 1, we specified the individual manufactures on which we conducted this study. In study 2, we conducted interviews based on the Evaluation Grid Method and extracted 10 positive emotions and 12 negative emotions that were evoked during individual manufacture. In study 3, we investigated the characteristics of the emotions along a valence axis and an arousal axis based on the core affect mode. These emotions had strong positive correlations between their valence and arousal. This correlation indicates the feedback functions of the emotions: the acceleratory function of positive emotions and the inhibitory function of the negative emotions. In addition, we divided these emotions into four clusters and found different patterns of evocation in the categories of manufacture and timing of the manufacture between clusters, which suggests the presence of qualitative differences.

View full abstract

In sports coaching, the coach often try linguistic instruction by collective secret language "jargons" used in the field. It is expected that the players can understand the meaning in jargons as interface and share a common concept through communication from the coach to the players. This research aims to explore establishment process of meaning in jargons about sprint from the perspective of the cognitive field. As a method, we analyzed the verbalization reports in which jargons about sprint from the players of the 42nd Japan national high school rugby team as its analytical resource by SCAT. These results suggest that jargons associated on firstly bodily sensation, and later associated with their behaviors about rugby．Meanwhile，it is also shown that such association have a correlation with the difficulty level of the motions of jargons．As a result of this consideration, the mechanism of symbolic communication was clarified, and furthermore, to design guidelines for linguistic instruction were obtained, for example, an effective coaching by jargons about motions easy to feel bodily sensation.

View full abstract

Elderly drivers generally are not good at recognizing multiple objects at the same time because they have less attention resources than younger drivers. As a result, when multiple hazards arise simultaneously while driving, elderly drivers may fail to promptly identify certain hazards, leading to an accident. This study examined, using a driving simulator, to what extent elderly people properly recognized multiple, simultaneous hazards. We also examined whether they recognized hazards more quickly if given a visual alert. The results showed that they could recognize hazards more quickly when the alert to a new hazard included directional information, and that, compared with younger people, elderly drivers tended to take their foot off the gas pedal while trying to identify the hazard. Further, a subjective assessment confirmed that they found the cognitive load of processing a directional alert harder than younger people. These findings suggest that optimizing the way the alert is given according to the severity and imminence of the hazard would make the alert more effective while reducing the driver’s cognitive load.

View full abstract

The aim of this study is to propose a method of evaluating for level of driver’s mental workload by using machine learning methods. In order to examine parameters related to visual behavior effective for evaluating driver's mental workload in driving, we measured EOG and head movements during simulated driving. In the experiment, we attempted to distinguish the "low mental workload" condition from "high mental workload" condition by using various machine learning methods, such as Adaboost, RBF networks and SVM on data related to EOG and head movements. Through cross-validation using the data from one participant as test data and data from the others as training data, the Adaboost method was determined to have the highest correct discrimination rate (over 85%). These results suggest the possibility of evaluating mental workload while driving.

View full abstract

Slide-in and Tap (SliT) is a software keyboard developed for smartwatches which focuses on reducing screen occupation. The initial display has three edges of the screen, excluding the bottom edge, each divided into two segments. One segment is selected using a slide-in motion, which is an operation of sliding a finger from the outside of the screen to the center of the screen. Given that the slide-in can be detected with a width of 2 mm, the screen occupancy of SliT is 26.4% of a 1.6-inch screen. This percentage is lower than competing methods. Two rows of the Japanese syllabary table are assigned to each segment. When the slide-in crosses a segment, the screen is split into two areas and the names of the rows assigned to that segment are displayed one by one. By lifting the finger up from within the area, the displayed row is selected. Then, the screen is divided into 9 squares in a 3x3 pattern and five hiragana characters from the selected row are displayed individually. A character is selected by tapping on it. It is simple to learn this operation. In the experiment, the average input speed of subjects after 5.3 min of use was 28.7 [CPM].

View full abstract