Since the emergence of ChatGPT, research on large language models (LLMs) has actively progressed across various fields. LLMs, pre-trained on vast text datasets, have exhibited exceptional abilities in understanding natural language and planning tasks. These abilities of LLMs are promising in robotics. In general, traditional supervised learning-based robot intelligence systems have a significant lack of adaptability to dynamically changing environments. However, LLMs help a robot intelligence system to improve its generalization ability in dynamic and complex real-world environments. Indeed, findings from ongoing robotics studies indicate that LLMs can significantly improve robots’ behavior planning and execution capabilities. Additionally, vision-language models (VLMs), trained on extensive visual and linguistic data for the vision question answering (VQA) problem, excel at integrating computer vision with natural language processing. VLMs can comprehend visual contexts and execute actions through natural language. They also provide descriptions of scenes in natural language. Several studies have explored the enhancement of robot intelligence using multimodal data, including object recognition and description by VLMs, along with the execution of language-driven commands integrated with visual information. This review paper thoroughly investigates how foundation models such as LLMs and VLMs have been employed to boost robot intelligence. For clarity, the research areas are categorized into five topics: reward design in reinforcement learning, low-level control, high-level planning, manipulation, and scene understanding. This review also summarizes studies that show how foundation models, such as the Eureka model for automating reward function design in reinforcement learning, RT-2 for integrating visual data, language, and robot actions in vision-language-action models, and AutoRT for generating feasible tasks and executing robot behavior policies via LLMs, have improved robot intelligence.

The goal of this study was to investigate the potential of wearable limb ballistocardiography (BCG) to enable cuff-less blood pressure (BP) monitoring, by investigating the association between wearable limb BCG-based pulse transit time (PTT) and BP. A wearable BCG-based PTT was calculated using the BCG and photoplethysmogram (PPG) signals acquired by a wristband as proximal and distal timing reference (called the wrist PTT). Its efficacy as surrogate of BP was examined in comparison with PTT calculated using the whole-body BCG acquired by a customized weighing scale (scale PTT) as well as pulse arrival time (PAT) using the experimental data collected from 22 young healthy participants under multiple BP-perturbing interventions. The wrist PTT exhibited close association with both diastolic (group average r = 0.79; mean absolute error (MAE) = 5.1 mmHg) and systolic (group average r = 0.81; MAE = 7.6 mmHg) BP. The efficacy of the wrist PTT was superior to scale PTT and PAT for both diastolic and systolic BP. The association was consistent and robust against diverse BP-perturbing interventions. The wrist PTT showed superior association with BP when calculated with green PPG rather than infrared PPG. In sum, wearable limb BCG has the potential to realize convenient cuff-less BP monitoring via PTT.

The objective of this study was to investigate the efficacy of pulse transit time (PTT)-pulse wave analysis (PWA) fusion in cuff-less blood pressure (BP) trend tracking based on the wearable wrist ballistocardiogram (BCG). We constructed PTT and 36 candidate BCG PWA features based on the BCG and photoplethysmogram (PPG) signals acquired at the wrist. We performed a model-based analysis to select 6 candidate BCG PWA features most sensitive to BP. We aggregated the 6 BCG PWA features into novel predictors of diastolic, systolic, and pulse BP (DP, SP, and PP) orthogonal to PTT by covariance-maximizing dimensionality reduction. Then, we evaluated and compared the efficacy of BCG PTT and BCG PTT-PWA fusion in tracking the trend of DP, SP, and PP. Unique innovations of this study, generalizable to a range of physiological signals beyond the BCG, are (i) the systematic selection of PWA features based on model-based analysis and (ii) the development of PWA-based BP predictors orthogonal to PTT. Using the experimental wrist BCG and PPG signals collected from 23 human subjects, we demonstrated that (i) BCG PTT-PWA fusion may significantly outperform BCG PTT; (ii) PTT may play the primary role in tracking BP while BCG PWA features may still have an independent and complementary value (especially in tracking PP); and (iii) model-based selection of BCG PWA features can enhance the robustness of BP trend tracking based on BCG PTT-PWA fusion.

Vasopressor infusion (VPI) is used to treat hypotension in an ICU. We studied compliance with blood pressure (BP) goals during VPI and whether a statistical model might be efficacious for advance warning of impending hypotension, compared with a basic hypotension threshold alert. Retrospective data were obtained from a public database. Studying adult ICU patients receiving VPI at submaximal dosages, we analyzed characteristics of sustained hypotension episodes (>15 min) and then developed a logistic regression model to predict hypotension episodes using input features related to BP trends. The model was then validated with prospective data. In the retrospective dataset, 102-of-215 ICU stays experienced >1 hypotension episode (median of 2.5 episodes per day in this subgroup). When trained with 75% of retrospective dataset, testing with the remaining 25% of the dataset showed that the model and the threshold alert detected 99.6% and 100% of the episodes, respectively, with median advance forecast times (AFT) of 12 and 0 min. In a second, prospective dataset, the model detected 100% of 26 episodes with a median AFT of 22 min. In conclusion, episodes of hypotension were common during VPI in the ICU. A logistic regression model using BP temporal trend features predicted the episodes before their onset.

This study investigates the potential of the limb ballistocardiogram (BCG) for unobtrusive estimation of cardiovascular (CV) parameters. In conjunction with the reference CV parameters (including diastolic, pulse, and systolic pressures, stroke volume, cardiac output, and total peripheral resistance), an upper-limb BCG based on an accelerometer embedded in a wearable armband and a lower-limb BCG based on a strain gauge embedded in a weighing scale were instrumented simultaneously with a finger photoplethysmogram (PPG). To standardize the analysis, the more convenient yet unconventional armband BCG was transformed into the more conventional weighing scale BCG (called the synthetic weighing scale BCG) using a signal processing procedure. The characteristic features were extracted from these BCG and PPG waveforms in the form of wave-to-wave time intervals, wave amplitudes, and wave-to-wave amplitudes. Then, the relationship between the characteristic features associated with (i) the weighing scale BCG-PPG pair and (ii) the synthetic weighing scale BCG-PPG pair versus the CV parameters, was analyzed using the multivariate linear regression analysis. The results indicated that each of the CV parameters of interest may be accurately estimated by a combination of as few as two characteristic features in the upper-limb or lower-limb BCG, and also that the characteristic features recruited for the CV parameters were to a large extent relevant according to the physiological mechanism underlying the BCG.

By virtue of its direct association with the cardiovascular (CV) functions and compatibility to unobtrusive measurement during daily activities, the limb ballistocardiogram (BCG) is receiving an increasing interest as a viable means for ultra-convenient CV health and disease monitoring. However, limited insights on its physical implications have hampered disciplined interpretation of the BCG and systematic development of the BCG-based approaches for CV health monitoring. In this study, a mathematical model that can predict the limb BCG in responses to the arterial blood pressure (BP) waves in the aorta was developed and experimentally validated. The validated mathematical model suggests that (i) the limb BCG waveform reveals the timings and amplitudes associated with the aortic BP waves; (ii) mechanical filtering exerted by the musculoskeletal properties of the body can obscure the manifestation of the arterial BP waves in the limb BCG; and (iii) the limb BCG exhibits meaningful morphological changes in response to the alterations in the CV risk predictors. The physical insights garnered by the analysis of the mathematical model may open up new opportunities toward next generation of the BCG-based CV healthcare techniques embedded with transparency, interpretability, and robustness against the external variability.