Research

As autonomous social robots become more prevalent in home environments, they must decide where to position themselves within many different types of rooms or spaces, balancing accessibility with staying out of the way. This paper presents a machine learning approach to modeling user preferences for robot parking spots in the home using standard 2D occupancy maps. Our method learns spatial patterns from the information available in the occupancy maps and user-annotated floorplans without requiring specialized inputs. We evaluate the approach using floorplan data from 84 users who provided parking spot preferences after living with and evaluating a social robot in their homes for at least two weeks. Our method significantly outperforms a state-of-the-art baseline focused exclusively on avoiding walking paths. We demonstrate how the approach extends to additional map features and share insights about the types of preference patterns learned by the model. This contribution provides a framework that can incorporate new environmental inputs as robot perception capabilities evolve.

For social robots to successfully integrate into daily life in home environments, they will need reliable models of the way people perceive and use space in the home. This paper explores the problem of obtaining annotated training data at scale for subjective judgments about spatial locations. Focusing on the use case of identifying good and bad parking spots for a social robot operating in a home environment, two experiments are presented. The first study shows that the presentation of context-rich 3D images to human annotators yields notably different outcomes from those obtained when using 2D robot navigation maps. We attribute the source of these differences to a set of features visible only in the 3D views and introduce a technique for labeling these features on the 2D maps. The second study reveals that using labeled 2D maps produces annotation data very similar to that obtained using 3D images. Since a labeled 2D map can be generated at a fraction of the cost of a full set of 3D views, we recommend this method as a scalable approach to collecting subjective spatial data annotations in everyday environments.

Current AI systems, including socially interactive robots, rely heavily on annotated datasets to train machine learning models. However, annotator biases can significantly affect the ground truth of these datasets, leading to biased AI behaviors. This paper investigates how the characteristics of the individuals comprising facial image datasets and online annotators may influence the reported perception of facial expressions. A between-subjects experiment with 119 online annotators revealed that Girls of Color were more prone to misclassification, even with similar facial expressions to White Girls. Annotators who identified as People of Color generally performed better at classifying expressions in Girls of Color, while White annotators were more accurate in classifying expressions in White Girls. Assistive AI-generated recommendations show potential in reducing racial bias, demonstrating a tangible approach toward fairness and accurate representation within ground truth data used to develop affective recognition systems. This work further describes critical implications for the development and deployment of socially interactive robots in diverse social contexts.

In recent news, organizations have been considering the use of facial and emotion recognition for applications involving youth such as tackling surveillance and security in schools. However, the majority of efforts on facial emotion recognition research have focused on adults. Children, particularly in their early years, have been shown to express emotions quite differently than adults. Thus, before such algorithms are deployed in environments that impact the wellbeing and circumstance of youth, a careful examination should be made on their accuracy with respect to appropriateness for this target demographic. In this work, we utilize several datasets that contain facial expressions of children linked to their emotional state to evaluate eight different commercial emotion classification systems. We compare the ground truth labels provided by the respective datasets to the labels given with the highest confidence by the classification systems and assess the results in terms of matching score (TPR), positive predictive value, and failure to compute rate. Overall results show that the emotion recognition systems displayed subpar performance on the datasets of children’s expressions compared to prior work with adult datasets and initial human ratings. We then identify limitations associated with automated recognition of emotions in children and provide suggestions on directions with enhancing recognition accuracy through data diversification, dataset accountability, and algorithmic regulation.

Humans can perceive multiple expressions, each one with varying intensity, in the picture of a face. We propose a methodology for collecting and modeling multidimensional modulated expression annotations from human annotators. Our data reveals that the perception of some expressions can be quite different across observers; thus, our model is designed to represent ambiguity alongside intensity. An empirical exploration of how many dimensions are necessary to capture the perception of facial expression suggests six principal expression dimensions are sufficient. Using our method, we collected multidimensional modulated expression annotations for 1,000 images culled from the popular ExpW in-the-wild dataset. As a proof of principle of our improved measurement technique, we used these annotations to benchmark four public domain algorithms for automated facial expression prediction.

The attribution of human-like characteristics onto humanoid robots has become a common practice in Human-Robot Interaction by designers and users alike. Robot gendering, the attribution of gender onto a robotic platform via voice, name, physique, or other features is a prevalent technique used to increase aspects of user acceptance of robots. One important factor relating to acceptance is user trust. As robots continue to integrate themselves into common societal roles, it will be critical to evaluate user trust in the robot’s ability to perform its job. This paper examines the relationship among occupational gender-roles, user trust and gendered design features of humanoid robots. Results from the study indicate that there was no significant difference in the perception of trust in the robot’s competency when considering the gender of the robot. This expands the findings found in prior efforts that suggest performance-based factors have larger influences on user trust than the robot’s gender characteristics. In fact, our study suggests that perceived occupational competency is a better predictor for human trust than robot gender or participant gender. As such, gendering in robot design should be considered critically in the context of the application by designers. Such precautions would reduce the potential for robotic technologies to perpetuate societal gender stereotypes.