There is significant interest in enabling robots to learn to perform tasks directly from interactions with non-expert users. Typically, a human serves as a teacher whose only task is to provide feedback to a robot learner. However, in real-world human-robot collaborations, the human often assists with the task while also offering feedback. Our key insight is that we can extract additional, implicit feedback from the human’s actions during the collaboration to augment the robot learning process. Under the assumption of fixed-role assignments, we first propose to formalize human preferences over a human-robot collaboration as a shared set of parameters encoding alignment between two reward functions: one that drives human behavior, and another that should direct robot behavior. This allows us to extract implicit feedback from an interaction by reasoning about the human’s actions in the task as actions that reveal the human’s preferences. Then, we combine this implicit feedback with traditional explicit human feedback to facilitate estimating the human’s preferences. We evaluated our proposed approach for Preference learning from Implicit and Explicit feedback (PIE) in simulations and with real users in a cooking scenario. Our simulation results indicate that combining multiple modalities of human feedback improves a robot’s ability to estimate human preferences over the collaboration, with a similar trend observed in real-world evaluations. These findings highlight a promising direction for enabling robots to adapt to a user’s preference model more quickly, thereby reducing the amount of time a person must spend teaching a robot.

Women comprise half of the population but continue to be underrepresented in STEM fields, especially engineering. This study explores the factors (individual, contextual, and barrier) influencing a woman's decision to pursue engineering as a college major and as a post-graduation career. A total of 47 questions covering 7 sections were developed in Google Forms and distributed via the Society of Women Engineers' social media platform plus Reddit's WomenEngineers page. Ninety-three people responded in total, with 86 classified as valid responses. Structural Equation Modeling (SEM) was used to analyze the relationships between the questions to underlying factors. Results from SEM found that interest in engineering is a significant factor in women's decisions to stay in the field. Additionally, the results indicated that, overall, there was no gender discrimination on campus; however, it was present in the workforce. ANOVA analyses revealed that there were significant differences in the perception of gender discrimination, work-life balance, and engineering choice among different demographic groups. Younger women reported perceiving less gender discrimination in the workplace compared to older women. Similarly, women who had been employed for less than one year perceived less gender discrimination than those with over five years of experience. Women working in environments where male colleagues were the majority reported higher levels of gender discrimination than those with a more balanced gender distribution among colleagues. In terms of work-life balance, women who had been employed for under a year reported better work-life balance than those who had been working for over a year. Additionally, women with undergraduate degrees in engineering were more likely to work in engineering fields than those who held a Ph.D. in engineering. In conclusion, this study highlights the significance of fostering interest in engineering, along with the impact of societal factors (such as gender discrimination policies) and individual factors (such as work-life balance skills), in supporting women pursuing careers in engineering.

Understanding human perceptions of robot performance is crucial for designing socially intelligent robots that can adapt to human expectations. Current approaches often rely on surveys, which can disrupt ongoing human–robot interactions. As an alternative, we explore predicting people’s perceptions of robot performance using non-verbal behavioral cues and machine learning techniques. We contribute the SEAN TOGETHER Dataset consisting of observations of an interaction between a person and a mobile robot in Virtual Reality, together with perceptions of robot performance provided by users on a 5-point scale. We then analyze how well humans and supervised learning techniques can predict perceived robot performance based on different observation types (like facial expression and spatial behavior features). Our results suggest that facial expressions alone provide useful information, but in the navigation scenarios that we considered, reasoning about spatial features in context is critical for the prediction task. Also, supervised learning techniques outperformed humans’ predictions in most cases. Further, when predicting robot performance as a binary classification task on unseen users’ data, the \(F_{1}\) -Score of machine learning models more than doubled that of predictions on a 5-point scale. This suggested good generalization capabilities, particularly in identifying performance directionality over exact ratings. Based on these findings, we conducted a real-world demonstration where a mobile robot uses a machine learning model to predict how a human who follows it perceives it. Finally, we discuss the implications of our results for implementing these supervised learning models in real-world navigation. Our work paves the path to automatically enhancing robot behavior based on observations of users and inferences about their perceptions of a robot.

This research-to-practice full paper compares data science education strategies in China and the United States, exploring whether different approaches can achieve similar educational outcomes. In the U.S., data science programs are typically developed by individual schools to meet labor market demands, integrating data science as an interdisciplinary subject. In contrast, Chinese universities follow a uniform plan set by the education department, resulting in specialized fields like ‘Big Data Management and Application’ and ‘Data Science and Big Data Technology.’ This research collected data on course offerings and university rankings in China and the U.S., analyzing curriculum content and program characteristics. Major findings indicate that in the U.S., Data Science and Business Analytics programs focus on technical skills and are primarily found in top-ranking universities, while in China, Data Science and Big Data Management programs emphasize management and are more widely distributed across different ranking groups. Despite these differences, both countries show similar geographic concentration patterns in data science programs. The study concludes that China and the U.S. adopt different educational strategies but achieve comparable effectiveness in data science education.

Recent work in Human-Robot Interaction (HRI) has shown that robots can leverage implicit communicative signals from users to understand how they are being perceived during interactions. For example, these signals can be gaze patterns, facial expressions, or body motions that reflect internal human states. To facilitate future research in this direction, we contribute the \textttREACT database, a collection of two datasets of human-robot interactions that display users' natural reactions to robots during a collaborative game and a photography scenario. Further, we analyze the datasets to show that interaction history is an important factor that can influence human reactions to robots. As a result, we believe that future models for interpreting implicit feedback in HRI should explicitly account for this history. \textttREACT opens up doors to this possibility in the future.

Numerous educational institutions, including those specializing in computer science, engineering, business, and science, now offer graduate programs in data science. However, it remains uncertain whether the data science programs provided by universities with different rankings deliver similar or distinct data science competencies. This study aims to compare the competencies of graduate data science programs offered in the United States. It addressed the research question of “Are there any differences in graduate data science program competencies based on their university rankings?” The program competencies of 228 data science graduate programs as well as their U.S. News university rankings were collected and compared. The findings indicate that regardless of their national rankings, all universities offer most of the competencies required for data science programs, except for Machine Learning and Sensor Networks. Forty-six percent of data science programs are offered in universities that hold national rankings within the top 75, focusing on majors such as data science and business analytics. Furthermore, the top-ranked universities provide data science programs across all departments or schools, with the most prominent ones being Business, Computer Science, Information Science & Technology, and Math/Statistics. The primary research contribution of this study revolves around conducting a comprehensive analysis of data science program competencies within the United States. The implications of the findings have significant relevance to computing education. The study provides valuable guidelines for educational practices in the field of data science: 1) naming the new majors as either “Data Science” or “Business Analytics”; 2) housing the new data science programs within existing departments or schools specializing in Business, Computer Science, Information Science & Technology, and Math/Statistics; 3) incorporating all ACM Data Science competencies into the curriculum of these programs while also emphasizing the need for a strong foundation in mathematics and statistics. Overall, these implications provide valuable guidance for universities seeking to develop impactful and well-rounded data science educational offerings that align with industry demands and equip students with the necessary skills and knowledge.

Recent research in robot learning suggests that implicit human feedback is a low-cost approach to improving robot behavior without the typical teaching burden on users. Because implicit feedback can be difficult to interpret, though, we study different methods to collect fine-grained labels from users about robot performance across multiple dimensions, which can then serve to map implicit human feedback to performance values. In particular, we focused on understanding the effects of annotation order and frequency on human perceptions of the self-annotation process and the usefulness of the labels for creating data-driven models to reason about implicit feedback. Our results demonstrate that different annotation methods can influence perceived memory burden, annotation difficulty, and overall annotation time. Based on our findings, we conclude with recommendations to create future implicit feedback datasets in Human-Robot Interaction.

Reactions such as gestures, facial expressions, and vocalizations are an abundant, naturally occurring channel of information that humans provide during interactions. An agent could leverage an understanding of such implicit human feedback to improve its task performance at no cost to the human. This approach contrasts with common agent teaching methods based on demonstrations, critiques, or other guidance that need to be attentively and intentionally provided. In this work, we demonstrate a novel data-driven framework for learning from implicit human feedback, EMPATHIC. This two-stage method consists of (1) mapping implicit human feedback to relevant task statistics such as reward, optimality, and advantage; and (2) using such a mapping to learn a task. We instantiate the first stage and three second-stage evaluations of the learned mapping. To do so, we collect a dataset of human facial reactions while participants observe an agent execute a sub-optimal policy for a prescribed training task. We train a deep neural network on this data and demonstrate its ability to (1) infer relative reward ranking of events in the training task from prerecorded human facial reactions; (2) improve the policy of an agent in the training task using live human facial reactions; and (3) transfer to a novel domain in which it evaluates robot manipulation trajectories. In the video, we focus on demonstrating the online learning capability of our instantiation of EMPATHIC.

The echo received by ground-to-air radar is a range-azimuth-pitch three-dimensional data. After pulse compression, the data of each range unit can be regarded as an azimuth-pitch two-dimensional (2D) echo. The resolution of azimuth and pitch is limited to antenna aperture. In this paper, the well-known Wiener filtering, Richardson-Lucy (RL) and total variation (TV) methods are introduced to simultaneously improve the azimuth-pitch resolution of ground-to-air radar. We first analyze the received signal of ground-to-air-radar, and model the echo of each range unit as a 2D convolution of target reflectivity distribution and azimuth-pitch antenna pattern. Then we deduce the Wiener filter, RL and TV methods in detail, and theoretically realize the super-resolution imaging of the azimuth and pitch. Finally, the super-resolution performance of different methods is verified by simulation.

Sparse regularization method has been widely used to realize super-resolution imaging in radar forward-looking imaging. However, most of existed methods directly minimize a nondifferentiable L1 regularization problem. In this paper, a Majorize-Minimization (MM) based super-resolution method is proposed to realize super-resolution for radar forward-looking imaging. According to MM principle, the proposed method converts the non-differentiable L1 regularization problem into a differentiable L2 regularization problem, and the real target distribution is obtained by solving the L2 regularization problem. Due to the introduction of the sparse prior, the proposed method can better improve the azimuth resolution of radar forward-looking imaging. In addition, the application of MM principle makes the non-differentiable L1 regularization easier to be solved. Finally, the superior performance of the proposed method is verified by simulation.

Total variation (TV) method has been utilized to realize super-resolution and preserve contour information of target in radar forward-looking imaging. However, its real-time ability is restricted to matrix inversion. In this paper, a fast TV (FTV) superresolution method is proposed to improve the real-time superresolution ability of traditional TV method. The proposed FTV method utilizes the low displacement rank features of Toplitz matrix and realizes fast matrix inversion by Gohberg-Semencul (GS) representation. It not only effectively improves the azimuth resolution and preserve the contour information of target, but also reduced the computational complexity of traditional TV method to improve its real-time superresolution ability. The superior performance of the proposed FTV method is verified by simulation and measured data processing.

Forward-looking imaging and target detection are highly desirable in many military and civilian fields, such as search and rescue, sea surface surveillance, airport surveillance, and guidance. However, there is a blind zone of forward-looking imaging for conventional Doppler beam sharpening and synthetic aperture radar. Scanning radar can be utilized to obtain a real beam image of a forward-looking area and implement target detection, while its azimuth resolution is poor due to the limitation of antenna size. Besides, during the processing procedure, imaging and target detection are usually regarded as two independent parts, which means that the imaging result will directly affect the detection performance. In this article, an integrated algorithm of super-resolution imaging and target detection for forward-looking scanning radar is proposed. In this algorithm, first of all, low-rank and sparse constraints as regularization norms are incorporated into the forward-looking scanning radar imaging and the objective function is established. Subsequently, the convex theory is utilized to solve the objective function and transform the problem of simultaneous super-resolution imaging and target detection into an optimization problem. Lastly, the super-resolution imaging and the target detection results are obtained simultaneously by solving the optimization problem using the alternating direction method of multipliers. In addition, simulation and experiment results are given to verify the effectiveness of the proposed algorithm.

Real-aperture radar can be utilized to realize forward-looking imaging by antenna scanning the imaging region. However, low azimuth resolution seriously affects its practical application. Although traditional super-resolution methods could enhance azimuth resolution to a certain extent, effective preservation of contour information for important targets still remains to be a problem. In this article, a method of total variation-sparse (TV-sparse) multiconstraint deconvolution is proposed to improve azimuth resolution of forward-looking imaging as well as preserve contour information of important targets. Since our interested targets usually appear to be sparse, the sparse constraint of the target is introduced first to achieve high resolution of forward-looking images, which may cause the loss of target contour information in the meantime. Second, total variation (TV) constraint is introduced based on the sparse constraint, converting traditional single-constraint super-resolution problem to a multiconstraint problem. We then use the split Bregman algorithm (SBA) to solve the multiconstraint problem, whose solution is the super-resolution image of radar forward-looking region. Compared with traditional super-resolution methods, the proposed method can improve the azimuth resolution of radar forward-looking imaging as well as better restore target contour information by adjusting respective weights of sparse constraint and TV constraint. Finally, the performance of the proposed method is validated with the simulation and measured data.

Target detection is an important function of forward-looking scanning radar search and tracking applications. However, it is difficult to detect the targets using the real beam image with low azimuth resolution. In this paper, a target detection scheme is proposed for forward-looking scanning radar. First, an image with better resolution is obtained by deconvolution technique, and it is used to map a patch-image. Then, according to the low rank characteristic of the patch-image and the sparse characteristic of the targets, the target detection is converted into an optimization problem of low rank and sparse matrix decomposition. Finally, the targets are obtained by solving this optimization problem. Simulations are given to verify its effectiveness.

The clear contour is required when realize azimuth superresolution of forward-looking radar imaging in many applications. Traditional deconvolution methods achieve the azimuth superresolution but are limited in contour recovery. Although the total variation (TV) method can be used to keep the contour information, it's sensitive to noise because of derivation. In this paper, we propose an improved total variation (ITV) method to realize azimuth superresolution of forward-looking radar imaging and recover the contour information. Firstly, the TV norm and L2norm are combined as the penalties under regularization framework. Then the regularization problem is solved by split Bregman algorithm. The proposed ITV method achieves higher azimuth resolution and better contour recovery performance than traditional methods, and the super performance is verified by simulations lastly.

In forward-looking scanning radar imaging, the azimuth resolution can be improved by adding the sparse constraint. However, the azimuth resolution is limited with noise influence by traditional sparse regularization methods. In this paper, we propose a Bayesian super-resolution method that solves the L1regularization problem using the split Bregman algorithm. This method decouples L1and L2norms for the independence of them to reduce the computational complexity. The simulations verify that the proposed algorithm provides a better resolution and de-noising ability compare with conventional methods.

This paper describes an empirical study that compared three alternatives for voice communication in conjunction with Web page collaboration for customer service.Two of the technologies used a single phone line for both voice and data transmission.These technologies were intemet telephony and Simultaneous< Voice and Data (SVD), a protocol which allows the voice to be routed over the public telephone network, rather hart the inteme~The study found that SVD was superior to intemet telephony in terms of a number of behavioral and subjective measures of con\7ersational interaction.The study also found that task time using intemet telephony was 45% greater than with SVD, mtilng the former a costly rdtemative in terms of human time.