Artificial intelligence (AI) and machine learning (ML) algorithms are increasingly being explored to support various decision-making tasks in health (e.g. rehabilitation assessment). However, the development of such AI/ML-based decision support systems is challenging due to the expensive process to collect an annotated dataset. In this paper, we describe the development process of a human-AI collaborative, clinical decision support system that augments an ML model with a rule-based (RB) model from domain experts. We conducted its empirical evaluation in the context of assessing physical stroke rehabilitation with the dataset of three exercises from 15 post-stroke survivors and therapists. Our results bring new insights on the efficient development and annotations of a decision support system: when an annotated dataset is not available initially, the RB model can be used to assess post-stroke survivor’s quality of motion and identify samples with low confidence scores to support efficient annotations for training an ML model. Specifically, our system requires only 22 - 33% of annotations from therapists to train an ML model that achieves equally good performance with an ML model with all annotations from a therapist. Our work discusses the values of a human-AI collaborative approach for effectively collecting an annotated dataset and supporting a complex decision-making task.

Advances in artificial intelligence (AI) have made it increasingly applicable to supplement expert’s decision-making in the form of a decision support system on various tasks. For instance, an AI-based system can provide therapists quantitative analysis on patient’s status to improve practices of rehabilitation assessment. However, there is limited knowledge on the potential of these systems. In this paper, we present the development and evaluation of an interactive AI-based system that supports collaborative decision making with therapists for rehabilitation assessment. This system automatically identifies salient features of assessment to generate patient-specific analysis for therapists, and tunes with their feedback. In two evaluations with therapists, we found that our system supports therapists significantly higher agreement on assessment (0.71 average F1-score) than a traditional system without analysis (0.66 average F1-score, p < 0.05). After tuning with therapist’s feedback, our system significantly improves its performance from 0.8377 to 0.9116 average F1-scores (p < 0.01). This work discusses the potential of a human-AI collaborative system to support more accurate decision making while learning from each other’s strengths.

A robotic exercise coaching system requires the capability of automatically assessing a patient's exercise to interact with a patient and generate corrective feedback. However, even if patients have various physical conditions, most prior work on robotic exercise coaching systems has utilized generic, pre-defined feedback.This paper presents an interactive approach that combines machine learning and rule-based models to automatically assess a patient's rehabilitation exercise and tunes with patient's data to generate personalized corrective feedback. To generate feedback when an erroneous motion occurs, our approach applies an ensemble voting method that leverages predictions from multiple frames for frame-level assessment. According to the evaluation with the dataset of three stroke rehabilitation exercises from 15 post-stroke subjects, our interactive approach with an ensemble voting method supports more accurate frame-level assessment (p <; 0.01), but also can be tuned with held-out user's unaffected motions to significantly improve the performance of assessment from 0.7447 to 0.8235 average F1-scores over all exercises (p <; 0.01). This paper discusses the value of an interactive approach with an ensemble voting method for personalized interaction of a robotic exercise coaching system.

Technology-assisted systems to monitor and assess rehabilitation exercises have an opportunity of enhancing rehabilitation practices by automatically collecting patient's quantitative performance data. However, even if a complex algorithm (e.g. Neural Network) is applied, it is still challenging to develop such a system due to patients with various physical conditions. The system with a complex algorithm is limited to be a black-box system that cannot provide explanations on its predictions. To address these challenges, this paper presents a hybrid model that integrates a machine learning (ML) model with a rule-based (RB) model as an explainable artificial intelligence (AI) technique for quantitative assessment of stroke rehabilitation exercises. For evaluation, we collected therapist's knowledge on assessment as 15 rules from interviews with therapists and the dataset of three upper-limb stroke rehabilitation exercises from 15 post-stroke and 11 healthy subjects using a Kinect sensor. Experimental results show that a hybrid model can achieve comparable performance with a ML model using Neural Network, but also provide explanations on a model prediction with a RB model. The results indicate the potential of a hybrid model as an explainable AI technique to support the interpretation of a model and fine-tune a model with user-specific rules for personalization.

Clinical decision support systems have the potential to improve work flows of experts in practice (e.g. therapist's evidence-based rehabilitation assessment). However, the adoption of these systems is challenging, and the gains of these systems have not fully demonstrated yet. In this paper, we identified the needs of therapists to assess patient's functional abilities (e.g. alternative perspectives with quantitative information on patient's exercise motions). As a result, we co-designed and developed an intelligent decision support system that automatically identifies salient features of assessment using reinforcement learning to assess the quality of motion and generate patient-specific analysis. We evaluated this system with seven therapists using the dataset from 15 patients performing three exercises. The results show that therapists have higher usage intent on our system than a traditional system without patient-specific analysis ($p < 0.05$). While presenting richer information ($p < 0.10$), our system significantly reduces therapists' effort on assessment ($p < 0.10$) and improves their agreement on assessment from 0.66 to 0.71 F1-scores ($p < 0.01$). This work discusses the importance of human centered design and development of a machine learning-based decision support system that presents contextually relevant information and salient explanations on its prediction for better adoption in practice.

Due to the limited number of therapists, task-oriented exercises are often prescribed for post-stroke survivors as in-home rehabilitation. During in-home rehabilitation, a patient may become unmotivated or confused to comply prescriptions without the feedback of a therapist. To address this challenge, this paper proposes an automated method that can achieve not only qualitative, but also quantitative assessment of stroke rehabilitation exercises. Specifically, we explored a threshold model that utilizes the outputs of binary classifiers to quantify the correctness of a movements into a performance score. We collected movements of 11 healthy subjects and 15 post-stroke survivors using a Kinect sensor and ground truth scores from primary and secondary therapists. The proposed method achieves the following agreement with the primary therapist: 0.8436, 0.8264, and 0.7976 F1-scores on three task-oriented exercises. Experimental results show that our approach performs equally well or better than multi-class classification, regression, or the evaluation of the secondary therapist. Furthermore, we found a strong correlation (R2 = 0.95) between the sum of computed exercise scores and the Fugl-Meyer Assessment scores, clinically validated motor impairment index of post-stroke survivors. Our results demonstrate a feasibility of automatically assessing stroke rehabilitation exercises with the decent agreement levels and clinical relevance.

Deep learning models have been successfully used in medical image analysis problems but they require a large amount of labeled images to obtain good performance. However, such large labeled datasets are costly to acquire. Active learning techniques can be used to minimize the number of required training labels while maximizing the model's performance. In this work, we propose a novel sampling method that queries the unlabeled examples that maximize the average distance to all training set examples in a learned feature space. We then extend our sampling method to define a better initial training set, without the need for a trained model, by using Oriented FAST and Rotated BRIEF (ORB) feature descriptors. We validate MedAL on 3 medical image datasets and show that our method is robust to different dataset properties. MedAL is also efficient, achieving 80% accuracy on the task of Diabetic Retinopathy detection using only 425 labeled images, corresponding to a 32% reduction in the number of required labeled examples compared to the standard uncertainty sampling technique, and a 40% reduction compared to random sampling.

Accurate real-time depth camera-based tracking of limbs, fingers and faces would be of great use to the field of Sign Language Recognition (SLR). While aspects of depth-based tracking have been applied to SLR, technological limitations have previously forced trade-offs between the resolution necessary to track finger positions and the field of view necessary to track the signer's body. Only recently, with improvements in cameras and computing power, have algorithms been developed which boast the capability of maintaining accurate finger tracking over an appropriately sized volume of space. In this paper, we employ the publicly available Sphere-Mesh [1] hand tracking algorithm to collect and recognize ASL handshapes. In doing so, we demonstrate recognition rates comparable to other state of the art handshape classifiers using simple naíve Bayesian classifiers that can run in real-time.

Given a retinal image, can we automatically determine whether it is of high quality (suitable for medical diagnosis)? Can we also explain our decision, pinpointing the region or regions that led to our decision? Images from human retinas are vital for the diagnosis of multiple health issues, like hypertension, diabetes, and Alzheimer's; low quality images may force the patient to come back again for a second scanning, wasting time and possibly delaying treatment. However, existing retinal image quality assessment methods are either black boxes without explanations of the results or depend heavily on feature engineering or on complex and error-prone anatomical structures' segmentation. Therefore, we propose EyeQual, that solves exactly this problem. EyeQual is novel, fast for inference, accurate and explainable, pinpointing low-quality regions on the image. We evaluated EyeQual on two real datasets where it achieved 100% accuracy taking just 36 milliseconds for each image.

In this work we present systems to monitor the health of a wide variety of high risk patients living with Congestive Heart Failure. For critical hospitalized patients, we introduce a deep learning framework for hospital records that uses a Word2Vec vector space representation to learn from a combination of structured data and unstructured text. The deep learning framework is able to assess patient risk, and accurately predict medical outcomes into the future. For less critical patients living at home, we also present algorithms for remote monitoring that can track a patient's changing health indicators using a wearable heart-rate sensors. The pool of individuals living with congestive heart failure is very diverse, which can make multifaceted approaches to health monitoring, such as those presented in this work, attractive for observing large pools of high-risk patients. Collectively the methods presented in this paper allow us to continuously monitor a person living with CHF through hospitalization, discharge, and into their home.

Though tactile maps have been shown to be useful tools for visually impaired individuals, their availability has been limited by manufacturing and design costs. In this paper, we present a system that uses 3D printing to (1) make tactile maps more affordable to produce, (2) allow visually impaired individuals to independently design and customize maps, and (3) provide interactivity using widely available mobile devices. Our system consists of three parts: a web interface, a modeling algorithm, and an interactive touchscreen application. Our web interface, hosted at www.tactilemaps.net, allows visually impaired individuals to create maps of any location on the globe while specifying (1) what features to map, (2) how the features should be represented by textures, and (3) where to place markers and labels. Our modeling algorithm accommodates user specifications to create map models with (1) multiple layers of continuously varying textures and (2) markers of various geometric shapes or braille characters. Our interactive application uses a novel approach to 3D printing tactile maps using conductive filament to provide touchscreen overlays that allow users to dynamically interact with the maps on a wide range of mobile devices. This paper details the implementation of our system. We also present findings from a user study validating the usability of our mapping interface and the utility of the maps produced. Finally, we discuss the limitations of our current implementation and the plans we have to improve our system based on feedback from our user study and additional interviews.

It is well established that delays in system response time negatively impact productivity, error rates and user satisfaction. What is less clear is the degree to which these effects deter users from engaging with a system. Usability guidelines provide rough response time targets for minimizing these effects across various types of interactions. However, developers faced with technical limitations or cost constraints that prevent them from meeting such targets are given no data with which to estimate the impact that system response delays will have on user engagement. In this work, we demonstrate a methodology for using crowd sourcing platforms to examine (1) the relative impacts of different delay types and (2) the effects of marginal changes in system response times. We compare two common network delay types, those caused by limited bandwidth (increased download times) and those caused by network latency (lag in responsiveness), and present how these delays reduce engagement in the context of a crowd sourced image classification task. Furthermore, we model how financial incentives interact with system response delays to impact user engagement. Finally, we show how such models can be used to optimize the cost of system design choices.

Sudden weight gain in patients living with Congestive Heart Failure (CHF) is often an indication that the individual is retaining fluid, which often means that patient's heart has weakened leading to increased risk of kidney or cardiac failure. Clinical interventions can be made at this stage, leading to better outcomes, however it is essential that the interventions take place before the patient's health declines too drastically. In this work, we present a latent variable autoregression model that tracks patient weight and blood pressure over time, allowing us to predict weight values into the future. We are also able to model continuous heart-rate signals and evaluate a subject's response to physical activity. This allows us to detect signs of health decline days earlier than existing rule-based systems, leading to the possibility of earlier clinical interventions, potentially preventing deadly medical emergencies.

This paper presents a method for filtering sinusoidal noise with a variable bandwidth filter that is capable of tracking a sinusoid's drifting frequency. The method, which is based on the adaptive noise canceling (ANC) technique, will be referred to here as the adaptive sinusoid canceler (ASC). The ASC eliminates sinusoidal contamination by tracking its frequency and achieving a narrower bandwidth than typical notch filters. The detected frequency is used to digitally generate an internal reference instead of relying on an external one as ANC filters typically do. The filter's bandwidth adjusts to achieve faster and more accurate convergence. In this paper, the focus of the discussion and the data is physiological signals, specifically electrocorticographic (ECoG) neural data contaminated with power line noise, but the presented technique could be applicable to other recordings as well. On simulated data, the ASC was able to reliably track the noise's frequency, properly adjust its bandwidth, and outperform comparative methods including standard notch filters and an adaptive line enhancer. These results were reinforced by visual results obtained from real ECoG data. The ASC showed that it could be an effective method for increasing signal to noise ratio in the presence of drifting sinusoidal noise, which is of significant interest for biomedical applications.

Tactile maps are useful, but not commonly available, tools for providing visually impaired individuals with knowledge about their environment. We have developed a web tool to allow visually impaired users to specify locations and customize 3D map models for production with 3D printers. Our tool uses available online map data and encodes features such as roads and waterways into 3D-printable tactile features. We present a preliminary overview of our web interface and tactile map-generating software with a focus on the design choices that were informed by our pilot studies. We will also discuss additional findings from our focus group interviews and future plans for this work.

In mobile computing, varying access to resources makes it difficult for developers to ensure that satisfactory system response times will be maintained at all times. Wearable physiological sensors offer a way to dynamically detect user frustration in response to increased system delays. However, most prior efforts have focused on binary classifiers designed to detect the presence or absence of a task-specific stimulus. In this paper, we make two contributions. Our first contribution is in identifying the use of variable length system response delays, a universal and task-independent feature of computing, as a stimulus for driving different levels of frustration. By doing so, we are able to make our second and primary contribution, which is the development of models that predict multiple levels of user frustration from psycho-physiological responses caused by system response delays. We investigate how incorporating different sensor features, application settings, and timing constraints impact the performance of our models. We demonstrate that our models of physiological responses can be used to classify five levels of frustration in near real-time with over 80% accuracy, which is comparable to the accuracy of binary classifiers.

The proliferation of wearable sensor platforms (Jawbone, Fit Bit, Smart Watches) and sensor rich Smart Phones have produced a torrent of real time sensed signals that is interpreted to produce data (e.g. Sleep quality, step count). Often the volume of this data overwhelms users. Visualization is an effective way to summarize data and observe trends. This is especially valuable when someone, such as a doctor, has to monitor a large number of people. When a human expert is not available, a Virtual Coach can provide feedback and guidance. Virtual coaches can recognize actions, correct errors, recognize emotions, and provide motivation. We will illustrate how these two technologies can effectively convert data into action in dozens of real applications. Examples include monitoring physiological parameters, identifying trends, providing guidance, correcting errors, and motivating. We will conclude with a projection of future Virtual Coaches.

Physical therapists could make better treatment decisions if they had accurate patient home exercise data but today this information is only available from patient self-report. A more accurate source of data could be gained from wearable computing designed for physical therapy exercise support. Existing systems have been tested in the lab but we have little information about issues they may face in home settings. We designed a technology probe, SenseCap, and deployed it for seven days in ten physical therapy patients' homes. SenseCap is a wearable physical therapy support system that gathers patient exercise compliance and performance data and summarizes the data in charts on an iPad Dashboard for physical therapists to view when patients return to the clinic. In this paper, we present the results of our deployment, show in-home patient exercise data gathered by the probe, and make design recommendations based on patient and physical therapist responses.

The paper presents an audio-based emotion recognition system that is able to classify emotions as anger, fear, happy, neutral, sadness or disgust in real time. We use the virtual coach as an application example of how emotion recognition can be used to modulate intelligent systems' behavior. A novel minimum-error feature removal mechanism to reduce bandwidth and increase accuracy of our emotion recognition system has been introduced. A two-stage hierarchical classification approach along with a One-Against-All (OAA) framework are used. We obtained an average accuracy of 82.07% using the OAA approach, and 87.70% with a two-stage hierarchical approach, by pruning the feature set and using Support Vector Machines (SVMs) for classification.

The combination of sensors, perception algorithms, and mobile computing enables situationally aware systems that provide proactive assistance. This paper outlines the basic components of a virtual coach with illustrations in five applications ranging from reminders to advice to opportunities for personal reflection.

RADAR is a multiagent system with a mixed-initiative user interface designed to help office workers cope with email overload. RADAR agents observe experts to learn models of their strategies and then use the models to assist other people who are working on similar tasks. The agents' assistance helps a person to transition from the normal email-centric workflow to a more efficient task-centric workflow. The Email Classifier learns to identify tasks contained within emails and then inspects new emails for similar tasks. A novel task-management user interface displays the found tasks in a to-do list, which has integrated support for performing the tasks. The Multitask Coordination Assistant learns a model of the order in which experts perform tasks and then suggests a schedule to other people who are working on similar tasks. A novel Progress Bar displays the suggested schedule of incomplete tasks as well as the completed tasks. A large evaluation demonstrated that novice users confronted with an email overload test performed significantly better (a 37% better overall score with a factor of four fewer errors) when assisted by the RADAR agents.

The six articles in this special issue focus on the technology and applications of body area networking.

Context-aware computing describes the situation where a wearable/mobile computer is aware of its user's state and surroundings and modifies its behavior based on this information. We designed, implemented, and evaluated a wearable system which can learn context-dependent personal preferences by identifying individual user states and observing how the user interacts with the system in these states. This learning occurs online and does not require external supervision. The system relies on techniques from machine learning and statistical analysis. A case study integrates the approach in a context-aware mobile phone. The results indicate that the method is able to create a meaningful user context model while only requiring data from comfortable wearable sensor devices.

Embedded systems encompass a wide range of applications, technologies, and disciplines, necessitating a broad approach to education. We describe embedded system coursework during the first 4 years of university education (the U.S. undergraduate level). Embedded application curriculum areas include: small and single-microcontroller applications, control systems, distributed embedded control, system-on-chip, networking, embedded PCs, critical systems, robotics, computer peripherals, wireless data systems, signal processing, and command and control. Additional cross-cutting skills that are important to embedded system designers include: security, dependability, energy-aware computing, software/systems engineering, real-time computing, and human--computer interaction. We describe lessons learned from teaching courses in many of these areas, as well as general skills taught and approaches used, including a heavy emphasis on course projects to teach system skills.

We at Carnegie Mellon University have pioneered context-aware mobile computing and built their first prototypes, including context-aware mobile phones and a context-aware personal communicator. These prototypes use machine learning and cognitive modeling techniques to derive user state and intent from the devices sensors. Context-aware computing describes the situation where a mobile computer is aware of its user's state and surroundings and modifies its behavior based on this information. We have demonstrated the power of our method to automatically derive a meaningful user context model and performed experimental measurements and evaluation. We have employed unsupervised machine learning techniques to combine real time data from multiple sensors into a model of behavior that is individualized to the user. We observe that context does not require a descriptive label to be used for adaptivity and contextually sensitive response. This makes our approach towards completely unsupervised machine learning feasible. By unsupervised learning we mean the identification of the users' context without requiring manually annotating current user states. We use unsupervised machine learning techniques to independently cluster sensor quantities and associate user interactions with these clusters. The use of this discretization enables learning from observations about the user. Each time a user interaction is observed, it is interpreted as a labeled example which can be used to construct a statistical model for context-dependent preferences. Example context-aware parameters are the following: location, nearby people and devices, calendar and other cyber sensors information, movement patterns and characteristics, user preferences, interests, and behavior patterns. By mapping observable parameters into cognitive states, the computing system can estimate the form of interaction that minimizes user distraction and the risk of cognitive overload. The capabilities herein proposed extend significantly the state-of-the-art, sometimes in a radical fashion, other times more incrementally. Our approach produces enriched observations by combining machine learning, instrumentation in software applications, sensors describing the user state, and task context information. Such diverse sensor fusion (symbolic and signal sensors) for inferring context and state goes much beyond the situation-sensing currently practiced, even in experimental settings.

THE convergence of a variety of technologies makes possible an entirely new way of using information processing. Continued advances in semiconductor technology produce high performance microprocessors requiring less power and less space. Decades of research in computer science have provided the technology for hands-free computing using speech and gesturing for input. Miniature heads-up displays weighing less than a few ounces have been introduced. Combined with mobile communication technology, it is possible for users to access information anywhere and anytime. Body-worn computers providing hands-free operation offer compelling advantages in many applications. Wearable computers deal in information rather than programs, becoming tools in the user’s environment much like a pencil or a reference book. The wearable computer provides portable access to information. Furthermore, the information can be automatically accumulated by the system as the user interacts with and modifies the environment, thereby eliminating the costly and errorprone process of information acquisition. Much as personal computers allow accountants and bookkeepers to merge their information space with their workspace (i.e., a sheet of paper), wearable computers allow mobile processing and the superposition of information on the users workspace. When combined with pervasive computing, wearable computers will provide access to the right information at the right place and at the right time. Distractions are even more of a problem when they occur in mobile environments than desktop environments since the user is often preoccupied with walking, driving, or other real-world interactions. A pervasive computing environment that minimizes distraction has to be context aware. Context-aware computing describes the situation where a mobile computer is aware of its user’s state and surroundings and modifies its behavior based on this information. A user’s context can be quite rich, consisting of attributes such as physical location, physiological state (such as body temperature, heart rate, and skin resistance), emotional state (such as angry, distraught, or calm), personal history, daily behavioral patterns, etc. If a human assistant were given such context, he or she would make decisions in a proactive fashion, anticipating user needs. In making these decisions, the assistant would typically not disturb the user at inopportune moments except in an emergency. The goal is to enable mobile computers to play an analogous role, exploiting context information to significantly reduce demands on human attention. Combined with inferences about users’ intentions, context-aware computing would allow improvement in user-perceived network and application performance and reliability. Context-aware intelligent agents can deliver relevant information when a user needs that information. These data make possible many exciting new applications, such as augmented reality, context aware collaboration, wearable assisted living, augmented manufacturing, and maintenance. Wearable computing brings the power of a pervasive computing environment to a person by placing computing and sensory resources on the user in an unobtrusive way. These computers can be specialized and modular, like items of clothing. Unlike laptops or handheld computers, wearable computers offer many new models to interact beyond keyboards and touch screens, in a natural, intuitive way, such as sound and tactile feedback. Also, wearables can be easily reconfigured to meet specific needs of applications. Every wearable computer system must be viewed from three different axes: the human, the computer, and the application. Within each of these axes there are difficult problems that must be solved and there are problems that arise from the fact that there are three axes. The human axis emphasizes wearability, which is defined as the interaction between the human body and the wearable object. Dynamic wearability includes the human body in motion. Design for wearability considers the physical shape of objects and their active relationship with the human form. Researchers explored history and cultures, including topics such as clothing, costumes, protective wearables, and carried devices. These studies of physiology, biomechanics, and movement were codified into guidelines for designing wearable systems. User comfort is a critical design consideration in many applications. New technologies such as smart textiles will significantly improve the functionality and ergonomics of wearable computers. The computer axis deals with the problems related to construction of a system with particular fabrics, size, power consumption, and user interface software. The application axis emphasizes mobile application design challenges and efficient mapping of problem solving capabilities to application requirements. Wearable computers have established their first foothold in several application domains, such as vehicle and aircraft maintenance and manufacturing, inspection procedures, augmented reality, context aware collaboration, language translation, etc. IEEE TRANSACTIONS ON COMPUTERS, VOL. 52, NO. 8, AUGUST 2003 977

This paper introduces a systematic approach to power awareness in mobile, handheld computers. It describes experimental evaluations of several techniques for improving the energy efficiency of a system, ranging from the network level down to the physical level of the battery. At the network level, a new routing method based upon the power consumed by the network subsystem is shown to improve power consumption by 15% on average and to reduce latency by 75% over methods that consider only the transmitted power. At the boundary between the network and the processor levels, the paper presents the problem of local versus remote processing and derives a figure of merit for determining whether a computation should be completed locally or remotely, one that involves the relative performance of the local and remote system, the transmission bandwidth and power consumption, and the network congestion. At the processor level, the main memory bandwidth is shown to have a significant effect on the relationship between performance and CPU frequency, which in turn determines the energy savings of dynamic CPU speed-setting. The results show that accounting for the main memory bandwidth using Amdahl's law permits the performance speed-up and peak power versus the CPU frequency to be estimated to within 5%. The paper concludes with a technique for mitigating the loss of battery energy capacity with large peak currents, showing an improvement of up to 10% in battery life, albeit at some cost to the size and weight of the system.

Summary form only given. Rapid advances in mobile computing and wireless communication present an opportunity for fundamental changes in engineering education, particularly for team based courses in which collaborative learning is the dominant learning mode. Through an iterative, bootstrapping approach, we study how students use computer-based collaboration tools to design a next generation of collaboration tools. Our suite of collaboration tools is called Handy Andy, which is based on the Andrew wireless network at Carnegie Mellon. Our goal is to facilitate computer-based collaborative learning in project-based design courses by developing collaboration tools for mobile computers.

This paper focuses on the design and implementation of a companion contextual car driver interface that proactively assists the driver in managing information and communication. The prototype combines a smart car environment and driver state monitoring, incorporating a wide range of input-output modalities and a display hierarchy. Intelligent agents link information from many contexts, such as location and schedule, and transparently learn from the driver, interacting with the driver only when it is necessary.

The design process used to produce an innovative computer system is presented.The computer system that resulted from the process uses a circular motif both for the user interface and the input device.The input device is a dial and the user interface is visually organized around the concept of a circle.The design process itself proceeded in the presence of a great many constraints and we discuss these constraints and how an innovative design was achieved in spite of the constraints.

The Wearable Computer Project is a testbed integratingresearch on rapid design and prototyping. Based onrepresentative examples from six generations of wearablecomputers, the paper focuses on the differences in rapidprototyping using custom design versus off-the-shelfcomponents. The attributes characterizing these two designstyles are defined and illustrated by experimentalmeasurements. The off-the-shelf approach required ten timesthe overhead, 30% more cost, fifty times the storage resources,20% more effort, five times more power, but 30% less effort toport software than the embedded approach.

The paper describes the evolution of an Interdisciplinary Concurrent Design Methodology (ICDM) and the metrics used to compare four generations of wearable computer artifacts produced by the methodology at each stage of ICDM's growth.The product cycle is defined, its phases, and the design information representation for each phase.Six generic axes of design activity are defined, and the concept of benchmarking a complete design methodology using these axes is introduced.In addition an approach for measuring design complexity is proposed.When applied to the four generations of the CMU wearable computers, the ICDM has demonstrated two orders of magnitude increase in design and efficiency. A. Conceptual Product.During the conceptualization stage, the multidisciplinary design team establishes a common vision of the end product.This vision provides a consistent set of design goals for all disciplines to maintain