Embodied agents must reason causally, as correlation-based models fail under intervention and distribution shift. This challenge arises in domains like robotics and cyber-physical systems, where agents balance efficiency and comfort under uncertainty. We introduce POLICYGRID, unifying causal discovery and control by treating each action as both decision and experiment. Leveraging constraint-based search, neural causal models, and language model priors with interventional validation, POLICYGRID yields adaptive, interpretable policies. Across synthetic, real-world, and live deployments, it achieves superior causal recovery (F1 = 0.89) and 2.8× better multi-objective performance than correlation-based baselines, demonstrating safe, generalizable decision-making.

Visual autoregressive modeling has recently demonstrated potential in image tasks by enabling coarse-to-fine, next-level prediction. Most indoor 3D occupancy prediction methods, however, continue to rely on dense voxel grids and convolution-heavy backbones, which incur high computational costs when applying such coarse-to-fine frameworks. In contrast, cost-efficient alternatives based on Gaussian representations—particularly in the context of multi-scale autoregression—remain underexplored. To bridge this gap, we propose DFGauss, a dynamic focused masking framework for multi-scale 3D Gaussian representation. Unlike conventional approaches that refine voxel volumes or 2D projections, DFGauss directly operates in the 3D Gaussian parameter space, progressively refining representations across resolutions under hierarchical supervision. Each finer-scale Gaussian is conditioned on its coarser-level counterpart, forming a scale-wise autoregressive process. To further enhance efficiency, we introduce an importance-guided refinement strategy that selectively propagates informative Gaussians across scales, enabling spatially adaptive detail modeling. Experiments on 3D occupancy benchmarks demonstrate that DFGauss achieves competitive performance, highlighting the promise of autoregressive modeling for scalable 3D occupancy prediction.

The Internet of Things (IoT) offers transformative potential when combined with Machine Learning (ML), but labeling diverse IoT data remains challenging. To address this, we introduce SenseScape Testbed, an IoT experimentation platform for indoor environments with wireless sensor nodes, robots, and location-tracking nodes. This testbed enables IoT applications such as human activity recognition and indoor mobility tracking, supporting energy efficiency, occupant comfort, and context representation while providing a versatile environment for labeling and testing to advance ML algorithms tailored for IoT applications.

Medication non-adherence is a major healthcare challenge globally, with over half of patients with chronic conditions in developed countries failing to follow their prescribed medication regimen. This can lead to poor disease outcomes, increased hospital visits, and a significant financial burden on healthcare systems [1]. These issues have driven a recent wave of research, including the development of smart adherence products [6] that can be incorporated into a patient’s daily life to monitor medication adherence. In this work, we present a radar-based system for user identification while taking medication, which extends our recent work [5]. we conducted preliminary experiments examining semi-medication-taking activities executed by 6 subjects. Our system achieved 80% accuracy in identifying who has taken the medication in a group of 3 subjects.

Cyber-physical-human systems (CPHS) in AI-based driver assistance applications require the integration of data from diverse modalities. These in-cabin CPHSs offer rich sensing capabilities, encompassing the vehicle, its surroundings, and the driver. One of the primary challenges in CPHSs is the incorporation of human behavior modeling to steer interactions that bolster human performance. This study introduces an in-cabin vehicular sensing framework that merges a driving simulator with a CPHS, thereby enabling researchers to devise and test multi-sensor fusion methodologies in human-in-the-loop driving situations. Initial experiments reveal that a driver interruptibility model can be effectively trained using the data gathered from our system.

We quantify the classification accuracy of Neural Networks (NNs) to specific websites using only the packet size and difference in inter-packet arrival time, which are easily observable via passive attackers in the network. Our flow classification work with NNs is unique in that we do not classify traffic by application type. Rather, we observe the accuracy of various NNs classifying specific web sites using HTTP traffic over TCP. We test a diverse set of neural network structures including a fully connected network (FCN), a convolutional neural network (CNN), a long short-term memory network (LSTM), and an autoencoder network (AE). We found that CNNs consistently had the highest accuracy, typically 80-90% when using 20 million packets as training data. We suspect that individual websites generate unique traffic patterns which are discoverable using NN techniques. Our work has important privacy implications. In particular, our work supports that both packet sizes and inter-packet timing must be randomized to obtain strong web browsing privacy. Many privacy preserving techniques, such as VPNs, will require additional enhancements.

A lack of sufficient situational awareness is a primary cause of traffic crashes due to human error. Redirecting a driver’s attention to critical objects is essential, but alerting driver about all critical objects can lead to distraction. This paper develops and evaluates an adaptive support system that incorporates drivers’ fixations as a proxy for their situational awareness. We implement an experimental system that detects a driver’s gaze on important objects in the traffic scene and adapts a cueing strategy in an augmented reality-based driver awareness assistance interface. We collect and analyze data from 15 participants and show that our adaptive support system strategy is effective without increasing the drivers’ cognitive workload. Finally, we show that our system can increase ratio of drivers’ fixations on critical objects in their view without significantly increasing dwell time per object.

Quality sleep is very important for a healthy life. Nowadays, many people around the world are not getting enough sleep, which has negative impacts on their lifestyles. Studies are being conducted for sleep monitoring and better understanding sleep behaviors. The gold standard method for sleep analysis is polysomnography conducted in a clinical environment, but this method is both expensive and complex for long-term use. With the advancements in the field of sensors and the introduction of off-the-shelf technologies, unobtrusive solutions are becoming common as alternatives for in-home sleep monitoring. Various solutions have been proposed using both wearable and non-wearable methods, which are cheap and easy to use for in-home sleep monitoring. In this article, we present a comprehensive survey of the latest research works (2015 and after) conducted in various categories of sleep monitoring, including sleep stage classification, sleep posture recognition, sleep disorders detection, and vital signs monitoring. We review the latest research efforts using the non-invasive approach and cover both wearable and non-wearable methods. We discuss the design approaches and key attributes of the work presented and provide an extensive analysis based on ten key factors, with the goal to give a comprehensive overview of the recent developments and trends in all four categories of sleep monitoring. We also collect publicly available datasets for different categories of sleep monitoring. We finally discuss several open issues and future research directions in the area of sleep monitoring.

Medication adherence is one of the leading factors that can make the difference between life and death, especially for patients managing chronic conditions [2]. Indeed, these issues have driven a recent wave of research, including the development of smart pill bottles that monitor when a pill is extracted. In this poster, we extend our recent work [1], where we present adaptive learning techniques for subject identification across multiple pill bottle systems. We collect inertial signals from 10 subjects taking medication pills and encode the activity signals by transforming them into 2D texture images. Then we use pre-trained Convolutional Neural Network (CNN) models for image-based classification tasks. Our approach achieved improved differentiation capacity over existing models by using deep learning models, modified through domain adaptation and transfer learning.

The COVID-19 pandemic has impacted academic life in different ways. In the mobile and pervasive computing community, there was a struggle on data collection for the evaluation of human-sensing systems. An automated and contactless solution to collect data from users at home is one way that can help in the continuation of user-centric studies. In this poster, we present a portable system for remote, in-home data collection. The system is powered by a Raspberry Pi© and input peripherals (a camera, a microphone, and a wireless receiver). Our system uses a speech interface for text-to-speech and speech-to-text conversions. The system acts as a voice-based "smart agent" that guides the user during an experiment session. We aim to use our system to collect data from a set of smart pill bottles that we previously designed for medication adherence monitoring [1] and user identification [3].

Supervised machine learning (ML) algorithms are aimed at maximizing classification performance under available energy and storage constraints. They try to map the training data to the corresponding labels while ensuring generalizability to unseen data. However, they do not integrate meaning-based relationships among labels in the decision process. On the other hand, natural language processing (NLP) algorithms emphasize the importance of semantic information. In this article, we synthesize the complementary advantages of supervised ML and NLP algorithms into one method that we refer to as SECRET (Semantically Enhanced Classification of REal-world Tasks). SECRET performs classifications by fusing the semantic information of the labels with the available data: it combines the feature space of the supervised algorithms with the semantic space of the NLP algorithms and predicts labels based on this joint space. Experimental results indicate that, compared to traditional supervised learning, SECRET achieves up to 14.0 percent accuracy and 13.1 percent F1 score improvements. Moreover, compared to ensemble methods, SECRET achieves up to 12.7 percent accuracy and 13.3 percent F1 score improvements. This points to a new research direction for supervised classification based on incorporation of semantic information.

Accurately accounting for medication use is important for the efficacy and safety of patients and family members. Monitoring is also important for medication adherence. This work investigates identification of persons taking medication using a sensor-equipped pill bottle. The bottle is equipped with inertial and switch sensors in both the cap and body, making the added hardware unobtrusive, low-cost, and wireless. Our system uses inertial data to build a patient discrimination model using classification techniques. We evaluated the system using 16 subjects. Our results show that using binary Support Vector Machine (SVM), the system can discriminate one patient among 16 subjects with 94% accuracy, and has a 93% using a single sensor. Identifying the exact person in a set of 3 subjects has an accuracy higher than 91%.

In this work, we investigate the identification of persons taking medication using a sensor-equipped pill-bottle. The bottle embeds inertial sensors in both the cap and body, making the added hardware un-obtrusive, low-cost, and wireless. Our system uses inertial data to build a patient discrimination model using classification techniques. We evaluated the system using 16 subjects. Our results show that using binary Support Vector Machine (SVM), the system can discriminate one patient among 16 subjects with 94 % accuracy. Identifying the exact person in a set of 3 subjects has an accuracy higher than 91 %..

In this paper we present deep learning based techniques for human activity classification that are designed to run in real time on mobile devices. Our methods minimize the size of the model and computational overhead in order to run on the embedded processor and preserve battery life. Prior work shows that the inertial measurement unit (IMU) data from waist-mounted mobile phones can be used to develop accurate classification models for various human activities such as walking, running, stair-climbing, etc. However, these models have largely been based on hand crafted features derived from temporal and spectral statistics. More recently, deep learning has been applied to IMU sensor data, but have not been optimized for resourceconstrained devices. We present a detailed study of the traditional hand-crafted features used for shallow/statistical models that consist of a over 561 manually chosen set of dimensions. We show, through principal component analysis (PCA) and application of a published support vector machine (SVM) pipeline, that the number of features can be significantly reduced - less than 100 features that give the same performance. In addition, we show that features derived from frequency-domain transformations do not contribute to the accuracy of these models. Finally, we provide details of our learning technique which creates 2D signal images from windowed samples of IMU data. Our pipeline includes a convolutional neural network (CNN) with several layers (1 convolutional layer and 1 averaging layer and a fully connected layer). We show that by removing the steps in the pipeline and layers in the CNN, we can still achieve 0.98 F1 score but with a much smaller memory footprint and corresponding computational cost. To increase the classification accuracy of our pipeline we added a hybrid bi-class support vector machine (SVM) that was trained using the labeled and flattened convolutional layer after each training image was processed. The learned feature set is almost half the size of the original hand crafted feature set and combining the CNN with the SVM results in 0.99 F1 score. We also investigate a novel application of transfer learning by using the time series 2D signal images to re-train two different publicly available networks, Inception/ImageNet and MobileNet. We find that re-trained ImageNet networks could be created $<; 5.5$ MB (suitable for mobile phones) and classification accuracy ranging from 0.83 to 0.93 (F1 score), thus indicating that retraining can be a useful future direction to build new classifiers for continuously evolving activities quickly while also being applicable to mobile device classification. Finally, we show that these deep learning models may be generalizable enough such that classifiers built from a given set of users for a specified set of activities can be used for a new user/subject as well.

In many event detection applications, training data may contain tags with multiple, simultaneous events. This is particularly likely when the definition of “event” is broad and includes events that can persist for an extended period of time. Decomposing a mixed signal into signals corresponding to individual events is non-trivial. In this paper, we propose a non-negative matrix factorization (NMF) method that generates independent dictionaries for different events from training data with overlapping events. The proposed method adds a mask matrix into the regularization term in conventional NMF approaches. This mask matrix captures known event labels in the training data, so that only related dictionary terms are updated during iteration. The effectiveness of the proposed approach is evaluated using both synthetic and real data.

Sensor network research has facilitated advancements in various domains, such as industrial monitoring, environmental sensing, etc., and research challenges have shifted from creating infrastructure to utilizing it. Extracting meaningful information from sensor data, or control applications using the data, depends on the metadata available to interpret it, whether provided by novel networks or legacy instrumentation. Commercial buildings provide a valuable setting for investigating automated metadata acquisition and augmentation, as they typically comprise large sensor networks, but have limited, obscure metadata that are often meaningful only to the facility managers. Moreover, this primitive metadata is imprecise and varies across vendors and deployments.

A typical large building contains thousands of sensors, monitoring the HVAC system, lighting, and other operational sub-systems. With the increased push for operational efficiency, operators are relying more on historical data processing to uncover opportunities for energy-savings. However, they are overwhelmed with the deluge of data and seek more efficient ways to identify potential problems. In this paper, we present a new approach called the Strip, Bind and Search (SBS); a method for uncovering abnormal equipment behavior and in-concert usage patterns. SBS uncovers relationships between devices and constructs a model for their usage pattern relative to other devices. It then flags deviations from the model. We run SBS on a set of building sensor traces; each containing hundred sensors reporting data flows over 18 weeks from two separate buildings with fundamentally different infrastructures. We demonstrate that, in many cases, SBS uncovers misbehavior corresponding to inefficient device usage that leads to energy waste. The average waste uncovered is as high as 2500~kWh per device.

We study the utility of dynamic frequency agility in real-world wireless sensor networks. Many view such agility as essential to obtaining adequate reliability in industrial environments. We quantify the actual utility by identifying the two facets of connectivity graphs that yield potential benefits called Multichannel Links (MCLs) and Multichannel Triangles (MCTs), study how frequently these occur empirically and determine whether multihop provides a comparable solution without the complexity of switching channels. We examine connectivity graphs of live networks over each 802.15.4 channel and find that MCLs and MCTs are extremely rare in practice. Almost no MCLs are found in any connectivity graph while MCTs occur between 0-200 parts per million (ppm). Furthermore, we show that MCLs are rarely important for routing while each MCT has a singlechannel routing solution. We also find that there are channels that are always good for connectivity and offer comparable routing costs, with respect to transmission count, in comparison to multichannel communication. Thus, the justification for channel agility in industrial environments applies in the absence but not in the presence of multihop routing.

As more and more physical information becomes available, a critical problem is enabling the simple and efficient exchange of this data. We present our design for a simple RESTful web service called the Simple Measuring and Actuation Profile (sMAP) which allows instruments and other producers of physical information to directly publish their data. In our design study, we consider what information should be represented, and how it fits into the RESTful paradigm. To evaluate sMAP, we implement a large number of data sources using this profile, and consider how easy it is to use to build new applications. We also design and evaluate a set of adaptations made at each layer of the protocol stack which allow sMAP to run on constrained devices.

As the number of wireless devices increase, the frequency spectrum becomes further congested. Deployments of wireless devices in harsh radio environments (i.e. an industrial plant) also motivates the study of alternate communication protocols that offer enough diversity to overcome interference. This work explores the use of frequency diversity to address this problem and examines its effectiveness in various environmental settings. We also examine the interplay between frequency agility at the MAC layer and route diversity in the network layer and look to understand the cost-tradeoffs in the diversity of choices offered by each layer.

In this paper we present Beacon Location Service (BLS): a location service for beacon-based routing algorithms like Beacon Vector Routing (BVR) [8] and S4 [19] . The role of a location service is to map node names to topologically meaningful addresses that can be used for routing. We evaluate an implementation of BLS that works on top of BVR. BLS resolves the destination node's name to BVR coordinates and then uses BVR to route the source message to the destination node.