People with physical impairments often face difficulties in performing daily tasks such as dressing, leading to dependence on caregivers. While robotic manipulators can provide valuable assistance, close physical interaction raises concerns over safety, comfort and trust, which can limit adoption. This article introduces the Assistive Robot Twin (ART) framework, a real-time digital twin that models both the user and the robot to enhance transparency during robot-assisted dressing. ART integrates two visual safety features: Bounding Boxes (BBs), which define static or dynamic protective zones around critical regions, and Trajectory Visualisation (TV), which displays planned robot movements in real time. We conducted a within-subject study with 36 participants mimicking stroke-related mobility impairment, evaluating six BB/TV configurations using validated interaction quality and system usability questionnaires. The results show that BBs significantly improved perceived safety ( \(\textrm{p} < 0.001\) ), reduced discomfort ( \(\textrm{p} < 0.001\) ) and increased trust ( \(\textrm{p} < 0.001\) ), with dynamic BBs providing the greatest safety benefits. TV significantly enhanced overall system usability ( \(\textrm{p}=0.007\) ), confidence and predictability of robot actions. While the study focuses on perceived interaction quality in a controlled setting with healthy participants, the results provide foundational evidence for the design of transparent assistive systems prior to clinical deployment.

As robotic systems become more interactive and complex, there is a need to standardise interfaces and simplify development processes. This is particularly pertinent in the field of manufacturing, where human-robot collaboration is on the increase, but where standards and proprietary software are key barriers to deployment and adoption.In this article we present the ROSIE Adapter, a general-purpose, modular adapter developed in ROS designed to support the creation and connection of industry-ready digital twins. Together with our previous work on the modular CSI digital-twin framework, we demonstrate how the ROSIE Adapter creates a versatile "plug-and-play" interface that simplifies the development of new robotic processes, and improves accessibility to novice users. Furthermore, the adaptor supports integration of intuitive interface devices, such as speech and augmented reality interfaces, which enable more natural collaboration. We describe the adaptor and its use in two real-world applications, demonstrate the ease of use via a three-day hackathon event, and provide results showing the faithfulness of the arising digital twins to their connected physical systems.

We present a tool-supported approach to the synthesis, verification, and testing of the control software responsible for the safety of human-robot interaction in manufacturing processes that use collaborative robots. In human-robot collaboration, software-based safety controllers are used to improve operational safety, for example, by triggering shutdown mechanisms or emergency stops to reduce the likelihood of accidents. Complex robotic tasks and increasingly close human-robot interaction pose new challenges to controller developers and certification authorities. Key among these challenges is the need to assure the correctness of safety controllers under explicit (and preferably weak) assumptions. Our integrated synthesis, verification, and test approach is informed by the process, risk analysis, and relevant safety regulations for the target application. Controllers are selected from a design space of feasible controllers according to a set of optimality criteria, are formally verified against correctness criteria, and are translated into executable code and tested in a digital twin. The resulting controller can detect the occurrence of hazards, move the process into a safe state, and, under certain circumstances, return the process to an operational state from which it can resume its original task. We show the effectiveness of our software engineering approach through a case study involving the development of a safety controller for a manufacturing work cell equipped with a collaborative robot.

Human-robot collaboration is on the increase and having a major impact on areas such as manufacturing, where the abilities of the human worker, augmented by those of the robot, bring increased flexibility and performance. However, close collaboration, including physical interaction, brings with it complex safety and security issues that were previously mitigated by human-robot segregation and isolated control networks. Exoskeletons pose a particularly interesting case whereby physical coupling of the user and robot is required throughout operation. We envisage the use of continuous authentication to exoskeletons, i.e. to ensure a user is who they claim to be, and that they have sufficient authority to operate the device for the duration of its use. In this paper we demonstrate such an approach to behavioural biometrics using data acquired through wearable sensors (hand manipulations recorded by a sensorised glove) while the user performs a selection of industrial tasks, including handling loads and inserting screws. The results show that the approach can discriminate between users with a low Equal Error Rate (EER; <3% in the worst case analysed). We believe that such an approach will also benefit other applications where wearables are used in robot control, such as in tele-operation.

Machine learning sits at the core of many essential products and services at Facebook. This paper describes the hardware and software infrastructure that supports machine learning at global scale. Facebook's machine learning workloads are extremely diverse: services require many different types of models in practice. This diversity has implications at all layers in the system stack. In addition, a sizable fraction of all data stored at Facebook flows through machine learning pipelines, presenting significant challenges in delivering data to high-performance distributed training flows. Computational requirements are also intense, leveraging both GPU and CPU platforms for training and abundant CPU capacity for real-time inference. Addressing these and other emerging challenges continues to require diverse efforts that span machine learning algorithms, software, and hardware design.

Many architects believe that major improvements in cost-energy-performance must now come from domain-specific hardware. This paper evaluates a custom ASIC---called a Tensor Processing Unit (TPU) --- deployed in datacenters since 2015 that accelerates the inference phase of neural networks (NN). The heart of the TPU is a 65,536 8-bit MAC matrix multiply unit that offers a peak throughput of 92 TeraOps/second (TOPS) and a large (28 MiB) software-managed on-chip memory. The TPU's deterministic execution model is a better match to the 99th-percentile response-time requirement of our NN applications than are the time-varying optimizations of CPUs and GPUs that help average throughput more than guaranteed latency. The lack of such features helps explain why, despite having myriad MACs and a big memory, the TPU is relatively small and low power. We compare the TPU to a server-class Intel Haswell CPU and an Nvidia K80 GPU, which are contemporaries deployed in the same datacenters. Our workload, written in the high-level TensorFlow framework, uses production NN applications (MLPs, CNNs, and LSTMs) that represent 95% of our datacenters' NN inference demand. Despite low utilization for some applications, the TPU is on average about 15X -- 30X faster than its contemporary GPU or CPU, with TOPS/Watt about 30X -- 80X higher. Moreover, using the CPU's GDDR5 memory in the TPU would triple achieved TOPS and raise TOPS/Watt to nearly 70X the GPU and 200X the CPU.

OBJECTIVE: This paper describes the University of Michigan's nine-year experience in developing and using a full-text search engine designed to facilitate information retrieval (IR) from narrative documents stored in electronic health records (EHRs). The system, called the Electronic Medical Record Search Engine (EMERSE), functions similar to Google but is equipped with special functionalities for handling challenges unique to retrieving information from medical text. MATERIALS AND METHODS: Key features that distinguish EMERSE from general-purpose search engines are discussed, with an emphasis on functions crucial to (1) improving medical IR performance and (2) assuring search quality and results consistency regardless of users' medical background, stage of training, or level of technical expertise. RESULTS: Since its initial deployment, EMERSE has been enthusiastically embraced by clinicians, administrators, and clinical and translational researchers. To date, the system has been used in supporting more than 750 research projects yielding 80 peer-reviewed publications. In several evaluation studies, EMERSE demonstrated very high levels of sensitivity and specificity in addition to greatly improved chart review efficiency. DISCUSSION: Increased availability of electronic data in healthcare does not automatically warrant increased availability of information. The success of EMERSE at our institution illustrates that free-text EHR search engines can be a valuable tool to help practitioners and researchers retrieve information from EHRs more effectively and efficiently, enabling critical tasks such as patient case synthesis and research data abstraction. CONCLUSION: EMERSE, available free of charge for academic use, represents a state-of-the-art medical IR tool with proven effectiveness and user acceptance.

Impact analysis - determining the potential effects of changes on a software system - plays an important role in software engineering tasks such as maintenance, regression testing, and debugging. In previous work, two new dynamic impact analysis techniques, CoverageImpact and PathImpact, were presented. These techniques perform impact analysis based on data gathered about program behavior relative to specific inputs, such as inputs gathered from field data, operational profile data, or test-suite executions. Due to various characteristics of the algorithms they employ, CoverageImpact and PathImpact are expected to differ in terms of cost and precision; however, there have been no studies to date examining the extent to which such differences may emerge in practice. Since cost-precision tradeoffs may play an important role in technique selection and further research, we wished to examine these tradeoffs. We therefore designed and performed an empirical study, comparing the execution and space costs of the techniques, as well as the precisions of the impact analysis results that they report. This paper presents the results of this study.

Impact analysis, determining when a change in one part of a program affects other parts of the program, is time-consuming and problematic. Impact analysis is rarely used to predict the effects of a change, leaving maintainers to deal with consequences rather than working to a plan. Previous approaches to impact analysis involving analysis of call graphs, and static and dynamic slicing, exhibit several tradeoffs involving computational expense, precision, and safety, require access to source code, and require a relatively large amount of effort to re-apply as software evolves. This paper presents a new technique for impact analysis based on whole path profiling, that provides a different set of cost-benefits tradeoffs - a set which can potentially be beneficial for an important class of predictive impact analysis tasks. The paper presents the results of experiments that show that the technique can predict impact sets that are more accurate than those computed by call graph analysis, and more precise (relative to the behavior expressed in a program's profile) than those computed by static slicing.

Impact analysis - determining the potential effects of changes on a software system - plays an important role in helping engineers revalidate modified software. In previous work we presented a new impact analysis technique. PathImpact, for performing dynamic impact analysis at the level of procedures, and we showed empirically that the technique can be cost-effective in comparison to prominent prior techniques. A drawback of that approach as presented, however, is that when attempting to apply the technique to a new version of a system as that system and its test suite evolves, the process of recomputing the data required by the technique for that version can be excessively expensive. In this paper, therefore, we present algorithms that allow the data needed by PathImpact to be collected incrementally. We present the results of a controlled experiment investigating the costs and benefits of this incremental approach relative to the approach of completely recomputing prerequisite data.