An official website of the United States government
Introduction As mobile robots proliferate in communities, designers must consider the impacts these systems have on the users, onlookers, and places they encounter. It becomes increasingly necessary to study situations where humans and robots coexist in common spaces, even if they are not directly interacting. This dataset presents a multidisciplinary approach to study human-robot encounters in an indoor apartment-like setting between participants and two mobile robots. Participants take questionnaires, wear sensors for physiological measures, and take part in a focus group after experiments finish. This dataset contains raw time series data from sensors and robots, and qualitative results from focus groups. The data can be used to analyze measures of human physiological response to varied encounter conditions, and to gain insights into human preferences and comfort during community encounters with mobile robots. Dataset Contents A dictionary of terms found in the dataset can be found in the "Data-Dictionary.pdf" Synchronized XDF files from every trial with raw data from electrodermal activity (EDA), electrocardiography (ECG), photoplethysmography (PPG) and seismocardiography (SCG). These synchronized files also contain robot pose data and microphone data. Results from analysis of two important features found from heart rate variability (HRV) and EDA. Specifically, HRV_CMSEn and nsEDRfreq is computed for each participant over each trial. These results also include Robot Confidence, which is a classification score representing the confidence that the 80 physiological features considered originate from a subject in a robot encounter. The higher the score, the higher the confidence A vectormap of the environment used during testing ("AHG_vectormap.txt") and a csv with locations of participant seating within the map ("Participant-Seating-Coordinates.csv"). Each line of the vectormap represents two endpoints of a line: x1,y1,x2,y2. The coordinates of participant seating are x,y positions and rotation about the vertical axis in radians. Anonymized videos captured using two static cameras placed in the environment. They are located in the living room and small room, respectively. Animations visualized from XDF files that show participant location, robot behaviors and additional characteristics like participant-robot line-of-sight and relative audio volume. Quotes associated with themes taken from focus group data. These quotes demonstrate and justify the results of the thematic analysis. Raw text from focus groups is not included for privacy concerns. Quantitative results from focus groups associated with factors influencing perceived safety. These results demonstrate the findings from deductive content analysis. The deductive codebook is also included. Results from pre-experiment and between-trial questionnaires Copies of both questionnaires and the semi-structured focus group protocol. Human Subjects This dataset contain de-identified information for 24 total subjects over 13 experiment sessions. The population for the study is the students, faculty and staff at the University of Texas at Austin. Of the 24 participants, 18 are students and 6 are staff at the university. Ages range from 19-48 and there are 10 males and 14 females who participated. Published data has been de-identified in coordination with the university Internal Review Board. All participants signed informed consent to participate in the study and for the distribution of this data. Access Restrictions Transcripts from focus groups are not published due to privacy concerns. Videos including participants are de-identified with overlays on videos. All other data is labeled only by participant ID, which is not associated with any identifying characteristics. Experiment Design Robots This study considers indoor encounters with two quadruped mobile robots. Namely, the Boston Dynamics Spot and Unitree Go1. These mobile robots are capable of everyday movement tasks like inspection, search or mapping which may be common tasks for autonomous agents in university communities. The study focus on perceived safety of bystanders under encounters with these relevant platforms. Control Conditions and Experiment Session Layout We control three variables in this study: Participant seating social (together in the living room) v. isolated (one in living room, other in small room) Robots Together v. Separate Robot Navigation v. Search Behavior A visual representation of the three control variables are shown on the left in (a)-(d) including the robot behaviors and participant seating locations, shown as X's. Blue represent social seating and yellow represent isolated seating. (a) shows the single robot navigation path. (b) is the two robot navigation paths. In (c) is the single robot search path and (d) shows the two robot search paths. The order of behaviors and seating locations are randomized and then inserted into the experiment session as overviewed in (e). These experiments are designed to gain insights into human responses to encounters with robots. The first step is receiving consent from the followed by a pre-experiment questionnaire that documents demographics, baseline stress information and big 5 personality traits. The nature video is repeated before and after the experimental session to establish a relaxed baseline physiological state. Experiments take place over 8 individual trials, which are defined by a subject seat arrangement, search or navigation behavior, and robots together or separate. After each of the 8 trials, participants take the between trial questionnaire, which is a 7 point Likert scale questionnaire designed to assess perceived safety during the preceding trial. After experiments and sensor removal, participants take part in a focus group. Synchronized Data Acquisition Data is synchronized from physiological sensors, environment microphones and the robots using the architecture shown. These raw xdf files are named using the following file naming convention: Trials where participants sit together in the living room [Session number]-[trial number]-social-[robots together or separate]-[search or navigation behavior].xdf Trials where participants are isolated [Session number]-[trial number]-isolated-[subject ID living room]-[subject ID small room]-[robots together or separate]-[search or navigation behavior].xdf Qualitative Data Qualitative data is obtained from focus groups with participants after experiments. Typically, two participants take part however two sessions only included one participant. The semi-structured focus group protocol can be found in the dataset. Two different research methods are applied to focus group transcripts. Note: the full transcripts are not provided for privacy concerns. First, we performed a qualitative content analysis using deductive codes found from an existing model of perceived safety during HRI (Akalin et al. 2023). The quantitative results from this analysis are reported as frequencies of references to the various factors of perceived safety. The codebook describing these factors is included in the dataset. Second, an inductive thematic analysis was performed on the data to identify emergent themes. The resulting themes and associated quotes taken from focus groups are also included. Data Organization Data is organized in separate folders, namely: animation-videos anonymized-session-videos focus-group-results questionnaire-responses research-materials signal-analysis-results synchronized-xdf-data Data Quality Statement In limited trials, participant EDA or ECG signals or robot pose information may be missing due to connectivity issues during data acquisition. Additionally, the questionnaires for Participant ID0 and ID1 are incomplete due to an error in the implementation of the Qualtrics survey instrument used.
more »
« less
Human Emotion Estimation through Physiological Data with Neural Networks
Effective collaboration between humans and robots necessitates that the robotic partner can perceive, learn from, and respond to the human's psycho-physiological conditions. This involves understanding the emotional states of the human collaborator. To explore this, we collected subjective assessments - specifically, feelings of surprise, anxiety, boredom, calmness, and comfort — as well as physiological signals during a dynamic human-robot interaction experiment. The experiment manipulated the robot's behavior to observe these responses. We gathered data from this non-stationary setting and trained an artificial neural network model to predict human emotion from physiological data. We found that using several subjects' data to train a general model and then fine-tuning it on the subject of interest performs better than training a model only using the subject of interest data.
more »
« less
- Award ID(s):
- 2326491
- PAR ID:
- 10552025
- Publisher / Repository:
- IEEE
- Date Published:
- Subject(s) / Keyword(s):
- Forward Transfer Robotics Human-Robot Collaboration
- Format(s):
- Medium: X
- Location:
- Tacoma, WA, USA
- Sponsoring Org:
- National Science Foundation
More Like this
-
-
Introduction:There is increasing interest in developing mathematical and computational models to forecast adverse events in physiological systems. Examples include falls, the onset of fatal cardiac arrhythmias, and adverse surgical outcomes. However, the dynamics of physiological systems are known to be exceedingly complex and perhaps even chaotic. Since no model can be perfect, it becomes important to understand how forecasting can be improved, especially when training data is limited. An adverse event that can be readily studied in the laboratory is the occurrence of stick falls when humans attempt to balance a stick on their fingertips. Over the last 20 years, this task has been extensively investigated experimentally, and presently detailed mathematical models are available. Methods:Here we use a long short-term memory (LTSM) deep learning network to forecast stick falls. We train this model to forecast stick falls in three ways: 1) using only data generated by the mathematical model (synthetic data), 2) using only stick balancing recordings of stick falls measured using high-speed motion capture measurements (human data), and 3) using transfer learning which combines a model trained using synthetic data plus a small amount of human balancing data. Results:We observe that the LTSM model is much more successful in forecasting a fall using synthetic data than it is in forecasting falls for models trained with limited available human data. However, with transfer learning, i.e., the LTSM model pre-trained with synthetic data and re-trained with a small amount of real human balancing data, the ability to forecast impending falls in human data is vastly improved. Indeed, it becomes possible to correctly forecast 60%–70% of real human stick falls up to 2.35 s in advance. Conclusion:These observations support the use of model-generated data and transfer learning techniques to improve the ability of computational models to forecast adverse physiological events.more » « less
-
null (Ed.)Recently, significant efforts are made to explore device-free human activity recognition techniques that utilize the information collected by existing indoor wireless infrastructures without the need for the monitored subject to carry a dedicated device. Most of the existing work, however, focuses their attention on the analysis of the signal received by a single device. In practice, there are usually multiple devices "observing" the same subject. Each of these devices can be regarded as an information source and provides us an unique "view" of the observed subject. Intuitively, if we can combine the complementary information carried by the multiple views, we will be able to improve the activity recognition accuracy. Towards this end, we propose DeepMV, a unified multi-view deep learning framework, to learn informative representations of heterogeneous device-free data. DeepMV can combine different views' information weighted by the quality of their data and extract commonness shared across different environments to improve the recognition performance. To evaluate the proposed DeepMV model, we set up a testbed using commercialized WiFi and acoustic devices. Experiment results show that DeepMV can effectively recognize activities and outperform the state-of-the-art human activity recognition methods.more » « less
-
null (Ed.)Abstract The increasing availability of passively observed data has yielded a growing interest in “data fusion” methods, which involve merging data from observational and experimental sources to draw causal conclusions. Such methods often require a precarious tradeoff between the unknown bias in the observational dataset and the often-large variance in the experimental dataset. We propose an alternative approach, which avoids this tradeoff: rather than using observational data for inference, we use it to design a more efficient experiment. We consider the case of a stratified experiment with a binary outcome and suppose pilot estimates for the stratum potential outcome variances can be obtained from the observational study. We extend existing results to generate confidence sets for these variances, while accounting for the possibility of unmeasured confounding. Then, we pose the experimental design problem as a regret minimization problem subject to the constraints imposed by our confidence sets. We show that this problem can be converted into a concave maximization and solved using conventional methods. Finally, we demonstrate the practical utility of our methods using data from the Women’s Health Initiative.more » « less
-
Keathley, H.; Enos, J.; Parrish, M. (Ed.)The role of human-machine teams in society is increasing, as big data and computing power explode. One popular approach to AI is deep learning, which is useful for classification, feature identification, and predictive modeling. However, deep learning models often suffer from inadequate transparency and poor explainability. One aspect of human systems integration is the design of interfaces that support human decision-making. AI models have multiple types of uncertainty embedded, which may be difficult for users to understand. Humans that use these tools need to understand how much they should trust the AI. This study evaluates one simple approach for communicating uncertainty, a visual confidence bar ranging from 0-100%. We perform a human-subject online experiment using an existing image recognition deep learning model to test the effect of (1) providing single vs. multiple recommendations from the AI and (2) including uncertainty information. For each image, participants described the subject in an open textbox and rated their confidence in their answers. Performance was evaluated at four levels of accuracy ranging from the same as the image label to the correct category of the image. The results suggest that AI recommendations increase accuracy, even if the human and AI have different definitions of accuracy. In addition, providing multiple ranked recommendations, with or without the confidence bar, increases operator confidence and reduces perceived task difficulty. More research is needed to determine how people approach uncertain information from an AI system and develop effective visualizations for communicating uncertainty.more » « less
- Free Publicly Accessible Full Text
-
Accepted Manuscript1.0
- Conference Paper:
- The DOI is not currently available.
