Find research datasets worth reusing

opencc-by-4.0Jan 2022View details →

HRI30: An Action Recognition Dataset for Industrial Human-Robot Interaction

A thorough analysis of the existing human action recognition datasets demonstrates that only a few HRI datasets are available that target real-world applications, all of which are adapted to home settings. Therefore, given the shortage of datasets in industrial tasks, we aim to provide the community with a dataset created in a laboratory setting that includes actions commonly performed within manufacturing and service industries. In addition, the proposed dataset meets the requirements of deep learning algorithms for the development of intelligent learning models for action recognition and imitation in HRI applications.

opencc-by-4.0Oct 2021View details →

Supplementary materials (set 2 of 2) in support of "Signalling Emotions with a Breathing Soft Robot" (Data set and materials used for human-robot interaction experiment)

Supplementary materials (set 2 of 2) in support of "Signalling Emotions with a Breathing Soft Robot" authored by Troels Aske Klausen, Ulrich Farhadi, Evgenios Vlachos, and Jonas Jørgensen. Contents of set 2:     - Data set and materials used for the human-robot interaction experiment and for data analysis Files:     - "Questionnaire.pdf": Questionnaire used for data collection.     - "Video links.txt": Weblinks to stimuli videos used.     - "Data set.xls": Collected raw data.     - "Matlab_DataAnalysis.mlx": Matlab script used to analyze raw data.     - "Linear_Arousal.png": Linear fit between the scoring of arousal and BPM.     - "Linear_Dominance.png": Linear fit between the scoring of dominance and BPM.     - "Linear_Pleasure.png": Linear fit between the scoring of pleasure and BPM. The experiment procedure is described in the paper. The soft robot used for the experiment is open source and can be manufactured using design files available on Zenodo: 10.5281/zenodo.5565201

opencc-by-sa-4.0Jul 2018View details →

Hand gesture dataset based on sEMG data captured from the Technaid human-robot interaction system

Two files with a dataset of five different/independent hand gestures are provided. The data were generated in a  sEMG system with two bracelets (eight sEMG sensors and six sEMG sensors) worn in the right forearm of a human. The Technaid human-robot interaction system was used to captured the data. The file "datasetForSegmentation.mat" was used to train a classifier whose purpose is the execution of Segmentation process. On the other hand, the file "datasetForRecognition.mat" was used to train a classifier whose purpose is the execution of gesture Recognition process.

opencc-by-sa-4.0Oct 2018View details →

Arm gesture dataset based on IMU data captured from the Technaid human-robot interaction system

Two files with a dataset of ten different/independent hand gestures are provided (seven static gestures and three dynamic gestures). The data were generated in a IMU system with five sensors worn in the right forearm, right arm, chest, left arm and left forearm of a human. The Technaid human-robot interaction system was used to captured the data. The file "datasetStaticGestures.mat" was used to train and test a classifier whose purpose is the recognition of static gestures. On the other hand, the file "datasetDynamicGestures.mat" was used to train and test a classifier whose purpose is the recognition of dynamic gestures. The latter file contains an extra class (gesture) which represents non-gestures.

opencc-by-4.0Feb 2024View details →

The AFFECT-HRI data set: physiological data for affective computing in human-robot interaction with anthropomorphic service robots

We provide a comprehensive data set AFFECT-HRI containing physiological data labeled with human affect (i.e., mood and emotion) gathered during an empirical study consisting of a complex human-robot interaction (HRI). A realistic retail scenario served as an experimental environment. In prior research, we showed the necessity to combine the expertise of the research fields of psychology, computer science, and law in the design of a responsible human-centered HRI. Therefore, we implemented five conditions (neutral, transparency, liability, moral, and immoral) covering the perspectives from these three research fields and used two different anthropomorphic service robots. Our study followed a multi-method approach, resulting in a data set containing and combining objective physiological sensor data with subjective human-affect assessments. Additionally, the data set includes insights from 146 participants regarding affect, demographics, and socio-technical questionnaire ratings, as well as robot gestures and robot speech. Our study can be split into three scenes: a consultation regarding products, a request for sensitive personal information while opening a customer account, and a successful or failing handover when buying a mold remover. Thus, this data set offers for the first time the possibility to prove established or develop new emotion recognition methods and technological capabilities for HRI. Further, our data set provides the possibility to combine affective computing with research about robot behavior (gestures, speech, and handover), liability (questionnaire), transparency (questionnaire), and psychological aspects, allowing an encompassing, human-centered view of HRI. The detailed data descriptor has been published in Nature Scientific Data. For more details on the data set, please check the paper below. Please cite the following paper if the dataset is used in a publication: Heinisch, J.S., Kirchhoff, J., Busch, P. et al. Physiological data for affective computing in HRI with anthropomorphic service robots: the AFFECT-HRI data set. Sci Data 11, 333 (2024). https://doi.org/10.1038/s41597-024-03128-z Acknowledgements This research was conducted as part of RoboTrust, a project of the Centre Responsible Digitality, supported by the Hessian Minister for Digital Strategy and Innovation. The authors would like to thank all participants for their participation in the study. We particularly want to thank Ruth Stock-Homburg for her support and for making Elenoide available. Further, we want to thank Mona Kegel, Vignesh Prasad, and all the research assistants who supported the study. We also thank the leap in time lab for serving as study location. A special thanks goes to Amer Altizini, who supported us by helping to prepare the data for publication. We want to thank Niklas Jungermann for his valuable comments on the statistical evaluation.

opencc-by-4.0Jul 2024View details →

Data and Code to Accompany: "On-Body Textile Hysteresis Estimation for Personalized Physical Human-Robot Interaction"

Data files, Matlab code, and figures to accompany "On-Body Textile Hysteresis Estimation for Personalized Physical Human-Robot Interaction" (submitted for peer-review on 7/20/2024).

Figure 11. Cognitive architecture of the process of social signals perception-Gestalt Processing in Human-Robot Interaction: A Novel Account for Autism Research

A possible cognitive architecture and formalization of the process of learning via multisensory integration is presented in figure 11. The formal description of the proposed cognitive architecture, capable of interpreting social-communication signals, signs and symbols, is based on multisensory integration at the level of perception, parallel processing at the level of interpretation and decision making followed by verbalization, as well as performing an action (eye contact, gesture, mimicking) at the level of behaviour.

Figure 9. The impossible figure (right) is not noticeable as such at first glance-Gestalt Processing in Human-Robot Interaction: A Novel Account for Autism Research

In the lexical domain a similar effect of holistic word processing is described in (Anstis, 2005b). The viewers were presented with pairs of three-letter words in quick succession and asked to report if the upper halves of the successively presented words were identical. Surprisingly, even when the upper halves of the words were orthographically identical, the error rate was reliably higher than expected and in comparison with matching identical successive words. As the author of the study Stuart Anstis points out: “students were processing the words not as separable parts, but holistically as perceptual units that could not be perceptually split apart. These results show that in normal circumstances, the visual system cannot, or does not, divide words into upper and lower halves” (Anstis, 2005b, p. 239).The author relates the results of his study to studies of visual perception of faces as evidence that the mechanism of holistic processing in the visual and the lexical domains is essentially the same.

Figure 8. Machine faces, perceived as more figure-like(left) and less figure-like (right)-Gestalt Processing in Human-Robot Interaction: A Novel Account for Autism Research

In an experimental study on visual perception, addressing directly Gestalt phenomena, a new Gestalt cue for figure-ground assignment was introduced (Vechera et al., 2002). The foreground versus the background organization is a strong determinant for decisions on objects seen among image elements. A well-known set of perceptual cues that are often called Gestalt cues are the size or area, the symmetry and the convexity vs. concavity judgments. It is generally assumed that figures are ‘small, symmetrical and convex’. The authors asked the question whether these cues are all that are necessary for a region of the image to be judged as a figure. The main result of this study is that regions in the lower portion of a stimulus array appear more figure-like than regions in the upper portion of the display.

Figure 6. Noticeable subjective response to the distorted face to the right-Gestalt Processing in Human-Robot Interaction: A Novel Account for Autism Research

Quite surprisingly, if the distortion is viewed in the normal upward position, it evokes strong emotional response to the distorted face to the right in figure 6.

Figure 5. The distortion is barely noticeable if the faces are viewed in the reversed position-Gestalt Processing in Human-Robot Interaction: A Novel Account for Autism Research

According to the feature-based processing theories of human faces the main elements, noticed and remembered in a face, are the eyes, the nose and the mouth (Thompson, 1980; Anstis, 2005a). If, however, we distort some of the elements of a face, these should influence perception,regardless of the position of the image – upright or reversed – from the observer viewpoint. Figure 5 presents the reversed image of the face on the left and the reversed distorted face on the right. The distortion was achieved by rotating the eyes of the image in the vertical direction. Figure

Figure 4. Main elements of a face, according to the feature-based processing theories-Gestalt Processing in Human-Robot Interaction: A Novel Account for Autism Research

In 1980 Peter Thompson proposed a new experimental paradigm for investigation of perception, called “face thatcherization” (also named “Thomson illusion”) (Thompson, 1980). Imagine that the following face, depicted in figure 4, is a photo of the then UK Prime Minister Margaret Thatcher.

Figure 3. Robotic faces, similar to smiley emoticons-Gestalt Processing in Human-Robot Interaction: A Novel Account for Autism Research

The smiley Gestalt is the result from a dynamic (evolved in time) cognitive process, it is immediately given in cognition, memorable, emotionally rich and socially relevant and reflects the special kind of Gestalt complexity as defined by Edwin Rausch (1988). Conventional representations of holistic entities like smileys or novel robotic faces come to life because they capture essential Gestalt qualities of the perceived image. For example, in figure 3 the robotic faces resemble the smileys in terms of the evoked internal/emotional reactions.

Figure 2. Taxonomy of the educational technologies for children with ASC-Gestalt Processing in Human-Robot Interaction: A Novel Account for Autism Research

learner (Perlin, 1997). MOSOCO is an emergent technology implemented in a smartphone called “Mobile Social Compass” (Escobedo et al., 2012). Six basic social skills are being encouraged by prompting the user to initiate social contact. The menu displays symbols for the basic social skills – eye contact, space and proximity, start an interaction, asking questions, sharing interests and finish an interaction. The MOSOCO application has turned out to be an extremely useful tool as an online prompt in starting, maintaining and finishing social interaction for both typical and autistic students, as well as to anyone that feels need for improving their social competence.

Figure 1. Necker cube depth illusion (Adapted from [http://en.wikipedia.org/wiki/Necker_cube])-Gestalt Processing in Human-Robot Interaction: A Novel Account for Autism Research

Often the term Gestalt is used interchangeably with the term “emergent whole” (Johansson, 1998). The emergence of a cognitive Gestalt structure adds dynamical and psychophysical forces, which are different from the static notion of the “emergent whole”. An eminent example for the dynamic nature of the emergent process is the Necker cube, which cannot be perceived as static, but rotates in front of our eyes to the complete exhaustion of the eye gazing process (figure 1).

Figure 7. Kanizsa square makes us see a non-existing figure – white square (Adapted from [http://en.wikipedia.org/wiki/Optical_illusion]-Gestalt Processing in Human-Robot Interaction: A Novel Account for Autism Research

A special case of Gestalt processing is the perceiving of illusions. Illusions make us see things or processes that are not there – for example the Kanizsa square like the one depicted in figure 7.