Jundi Liu

As robots are rapidly entering new aspects of human life, it is essential to understand how humans trust them while ensuring appropriate safety guarantees from the outset. However, during the early phases, identifying the full range of trust-driven behaviors humans may exhibit in interactions with robots remains challenging due to limited data. Therefore, capturing high-level similarities and differences among individuals can help robots acclimate to individual users over time. Supporting this process raises an important question: what type of data can better support such acclimation—subjective trust ratings, trust-driven physiological data, or a combination of both? To address this question, a human subject study is conducted in a virtual manufacturing environment where participants interact with Automated Guided Vehicles (AGVs) during task execution. From these interactions, patterns of trust-driven behaviors and subjective trust ratings are independently extracted. Behavioral patterns reveal two distinct clusters, termed Skeptical and Deliberate, while subjective ratings yield two groups corresponding to high and low trust. It was found that behavioral and subjective clustering produce fundamentally different group distinctions, indicating limited correspondence between the two. Moreover, differences in behavioral clusters correspond to significantly different crossing orders, i.e., whether the participant crosses before the AGV or yields to it. At the same time, such distinctions are only statistically significant when the user crosses first under subjective clustering. Finally, high-level design implications for transport robots have been outlined to help them gradually adapt their behavior to different groups of people as they gather more data to build better, personalized models.

Despite growing curricular emphasis on computational thinking (CT) and engineering design practice (EDP), preparing elementary teacher candidates (ETCs) for integrated, practice-based instruction remains challenging. Grounded in Communities of Practice (CoP) and design-based implementation research (DBIR) frameworks, this study examines how ETCs (pre-survey: n = 23, post-survey: n = 13) develop learning-to-teach practices for integrating coding and robotics into PreK–5 STEAM learning (STEM with Arts integration) to foster these competencies. The study is situated within an interdisciplinary, community-partnered course, Toying with Technology (TwT), where ETCs collaborate in Critical Friends teams to design, facilitate, and reflect on community partners’ need-oriented, empathy-centered STEAM experiences. Utilizing PreK–5 learning technologies and field-based insights from STEM/STEAM professionals, ETCs developed 5E lesson plans integrating science, literacy, mathematics, and robotics, culminating in both an in-person and a virtual STEAM Family Night. Data sources include pre- and post-surveys measuring ETCs’ technology integration perceptions, post-field exit tickets, family feedback, community partner reflections, and course artifacts. Pre-survey results indicated limited prior experience and neutral confidence regarding technology integration. In contrary, post-survey findings demonstrated significant increases in self-reported confidence, conceptual understanding of CT and EDP, and readiness to facilitate empathy-centered, technology-rich learning. Qualitative analysis further revealed growth in collaboration, problem-solving, and pedagogical reasoning, while parent and partner feedback highlighted high levels of engagement and increased STEAM interest. Overall, this study provides initial empirical evidence, within a community-partnered, empathy-oriented course design, showing how community-engaged teacher education, informed by CoP and DBIR frameworks, prepares ETCs to teach CT, EDP, and critical problem-solving through integrated, authentic STEAM integration.

One potential application for increasingly capable virtual reality and humanoid robots is enhancing classroom engagement, problem-solving, and social-emotional learning. To evaluate future educators’ readiness to implement these technologies, 25 future elementary teachers were surveyed on their attitudes towards robots, generalized self-efficacy, social and emotional learning competencies, and attitudes towards virtual reality and robots in education. Results indicate ambivalent personal level attitudes towards robots and negative societal level attitudes toward robots. Respondents rated their competencies highly, but their attitudes towards VR and robots in the classroom suggest that while they see potential in these technologies, they do not feel adequately informed or prepared to implement them or see robots as empathetic. Future teachers do not feel an immediate need for these technologies and would require appropriate training and support. Future work should explore specific uses of these technologies, their effects on classrooms, and the nature of support teachers would need.

Adaptive systems aim to address the human-out-of-the-loop problem in automation, associated with reduced situation awareness and users' overreliance. These systems modify their behavior based on system, task, or operator-based triggers, with the latter offering a human-centered approach by responding to different human states such as mental workload or fatigue. Despite growing interest in adaptive systems, no review has consolidated empirical findings. To fill this gap, and using PRISMA methodology, 34 articles were included after full-text screening. The analysis focuses on the types of human states measured, the sensors and modeling techniques used, the nature of the adaptive interventions, and their reported outcomes. Cognitive states were the most commonly assessed, often using EEG due to its ability to capture brain activity in real time. Physical state, frequently applied in rehabilitation, relied on EMG and IMU sensors to assess muscle activation and joint accuracy. Machine learning methods were more prevalent than rule-based systems, likely due to their flexibility and robustness. While function allocation emerged as the dominant adaptive strategy, system performance outcomes varied. Future work should prioritize user-centered design, avoiding task disruption or disengagement. Finally, standardized evaluation frameworks are required to assess human–machine performance.

Vehicle automation seeks to enhance road safety and improve the driving experience. However, a standard system does not account for variations in users and driving conditions. Customizing vehicle automation based on users’ preferences aims to improve the user experience and adoption of the technologies. This study introduces a systematic paradigm that starts with naturalistic driving data to identify the driving behaviors and styles for a customized automated lane change system. The driving behaviors are first extracted using Multivariate Functional Principal Component Analysis (MFPCA) with minimum prior expert knowledge. The driving styles are identified by clustering the extracted driving behaviors. An Inverse Reinforcement Learning (IRL) algorithm is then used to train the automated lane change system from grouped demonstrations of the identified driving styles to capture the preferences of a group of drivers with a similar driving style. The performance of the proposed customized automated lane change system is compared to (1) a non-customized system trained on all the sample trips, (2) customized systems built on expert-coded reward functions, and (3) customized systems trained using a Generative Adversarial Imitation Learning (GAIL) algorithm. The results show that our method outperforms all the other systems with respect to the prediction accuracy of the lane change actions. Additionally, our method gains insights on the representative behaviors of different driving styles to enable customization of automated lane change systems.

In current supply chain operations, original equipment manufacturers (OEMs) procure parts from hundreds of globally distributed suppliers, which are often small- and medium-scale enterprises (SMEs). The SMEs also obtain parts from many other dispersed suppliers, some of whom act as sole sources of critical parts, leading to the creation of complex supply chain networks. These characteristics necessitate having a high degree of visibility into the flow of parts through the networks to facilitate decision making for OEMs and SMEs, alike. However, such visibility is typically restricted in real-world operations due to limited information exchange among the buyers and suppliers. Therefore, we need an alternate mechanism to acquire this kind of visibility, particularly for critical prediction problems, such as purchase orders deliveries and sales orders fulfillments, together referred as work orders completion times. In this paper, we present one such surrogate mechanism in the form of supervised learning, where ensembles of decision trees are trained on historical transactional data. Furthermore, since many of the predictors are categorical variables, we apply a dimension reduction method to identify the most influential category levels. Results on real-world supply chain data show effective performance with substantially lower prediction errors than the original completion time estimates. In addition, we develop a web-based visibility tool to facilitate the real-time use of the prediction models. We also conduct a structured usability test to customize the tool interface. The testing results provide multiple helpful suggestions on enhancing the ease-of-use of the tool.

Multiagent systems are characterized by decentralized decision-making by the (semi)autonomous agents and localized communication or information exchange among the neighboring agents. Supply-demand networks form the backbones of both services and manufacturing industries, and need to operate as efficiently as possible to yield optimized returns. In this letter, we bring the notion of multiagent systems to clustered supply-demand networks such that each supplier acts as an agent. Consequently, we adapt consensus-based auction bidding methods to optimize the assignment of demands to the suppliers with known communication pathways and resource constraints. Results on moderately large networks show promising performance in terms of both assignment quality, as given by the overall demand delivery cost and proportion of assigned demands, and computation time.