Search for: All records

Fully decentralized model training for on-road vehicles enables leveraging crowdsourced data without relying on central servers, infrastructure, or persistent Internet connectivity. However, real-world vehicular scenarios pose fundamental challenges to decentralized learning, including highly dynamic network topology, unreliable wireless links, and heterogeneous radio capability constraints. To address these challenges, we proposeRoADTrain, a route-assisted decentralized peer model training framework with formal convergence guarantees. In RoADTrain, vehicles share short-horizon route information to estimate inter-vehicle contact duration and link reliability, which are embedded into a base graph. Vehicles then select communication partners by maximizing the algebraic connectivity of the resulting communication graph, thereby accelerating information mixing and model convergence. We further investigate subcarrier reuse and extend the algorithm to three representative radio capability cases. Extensive evaluations show that RoADTrain achieves convergence comparable to communication-unconstrained state-of-the-art approaches while reducing communication overhead by up to 4.5×, and significantly outperforms communication-constrained decentralized baselines by up to 29% higher driving success rates in online evaluations. We further empirically demonstrate the robustness of RoADTrain under a range of unexpected effects encountered in complex real-world applications, as well as in a large-scale traffic scenario. 

Probing and understanding ultrahigh-frequency/gigahertz (UHF/GHz) vibrations in atomic layer nanomechanical resonators holds strong promise for fundamental studies and technological applications, such as sensing, signal processing, and quantum engineering. However, accessing GHz flexural-mode resonances in such devices at room temperature has been challenging. Here, we demonstrate the first measurement of GHz flexural vibrations in graphene and molybdenum disulfide (MoS₂) resonators at room temperature, achieving multimode resonances (f_m,n) up to ~1.09 GHz and quality factors (Q_m,n) up to ~5400 in multilayer MoS₂resonators, andf_m,nup to ~1.03 GHz with highQ_m,n ~4500 in few-layer graphene resonators. Bothfs andQs are the highest among reported atomic layer nanomechanical resonators at room temperature, yieldingf_m,n × Q_m,n ~ 5 × 10¹²Hz. We also reveal aQscaling law with higher modes, which favors detecting GHz resonances. This study will enable exploiting multiple modes in atomic layer resonators toward resonant sensing and transduction functions at UHF/GHz. 

Introduction:More than 50% of stroke patients experience limitations in activities of daily living (ADLs). Poor performance in ADLs has been associated with increased risk of rehospitalization after stroke and worsened quality of life. Monitoring and measuring ADLs after stroke can provide valuable insight into patients’ recovery and help identify targeted therapies to improve overall functioning. The objective of this study was to determine the feasibility of using contactless room-based sensors to capture physiological parameters and movement in patients following a stroke and whether these data can identify differences in function compared to individuals with varying ages and health status. Methods:Individuals age 18 and older were recruited to come to a study one-bedroom apartment outfitted with 16 ultrawideband contactless sensors collecting physiological data and movement following a scripted activities of daily living (ADL/IADL) protocol. A study-developed app guided patients through scripted ADLs while measuring the time to completion of each activity. Activities were grouped into simple (brush teeth, wash face, e.g.) and complex activities (food preparation, manage laundry, e.g.). Mean time to completion of all activities and activity groups were compared across 5 participant groups: students, non-student adults, healthy older adults, acute stroke survivors, and chronic stroke survivors. Results:A total of 50 participants completed at least one study visit with an average completion time of 2034 seconds (s). There were significant differences between the time to completion of the visit, ranging from 1772s to 2516s (p<0.05). Stroke survivors required the greatest amount of time (2516s), and the acute stroke survivor took the shortest amount of time to complete the tasks (1771s) followed by the students (1809s). While there was no significant difference between participant groups regarding mean total time to complete simple tasks, there was a significant difference (p<.05) in mean total time to complete complex tasks with chronic stroke survivor taking the longest (242s). Conclusions:These data demonstrate the benefits of monitoring stroke recovery and the potential challenges that long-term stroke survivors have in completing ADLs/IADL. Long term, these data show the feasibility of evaluating differences between ADL/IADL performance and the potential to utilize such data to monitor changes in health status as patients recover from stroke. 

Abstract As Americans live longer, there is a growing need to develop and implement technologies to monitor changes in health, supporting aging in place. We have installed privacy-preserving ultra-wideband wireless sensors in a model home to facilitate sensor and algorithm development for future home deployment. These sensors monitor vital signs and provide localization/tracking and activity recognition. We have developed a protocol to continuously collect and measure activities of daily living, location, and gait. We developed a multi-modal app using videos, audio, and texts to guide individuals and track activity completion time. To protect data privacy, raw sensor data is sent to a research cloud, and identifiable image data is sent to a university-supported HIPAA-compliant Box app. We spent >10 months developing, evaluating, stress testing, and improving our system and cloud infrastructure stability to ensure <1% data loss. We completed 19 sessions with volunteers to test our app and protocol, provide feedback, and continue to update our procedures and app before beginning data collection with individuals post-stroke. Recovery progress in post-stroke study participants will complete 3 study visits over 14 weeks to capture clinical and sensor measurements. We continue to refine our home deployment protocols for older adults with diverse technological knowledge. This presentation highlights our progress in developing and deploying our home monitoring system. We will discuss the challenges and procedures to collect, analyze, and interpret continuous physiologic and movement data in real-time, providing information, knowledge, and insights that will become vital to supporting the health of our expanding aging population.