Wearable sensors have become ubiquitous thanks to a variety of health tracking features. The resulting continuous and longitudinal measurements from everyday life generate large volumes of data. However, making sense of these observations for scientific and actionable insights is non-trivial. Inspired by the empirical success of generative modeling, where large neural networks learn powerful representations from vast amounts of text, image, video, or audio data, we investigate the scaling properties of wearable sensor foundation models across compute, data, and model size. Using a dataset of up to 40 million hours of in-situ heart rate, heart rate variability, accelerometer, electrodermal activity, skin temperature, and altimeter per-minute data from over 165,000 people, we create LSM, a multimodal foundation model built on the largest wearable-signals dataset with the most extensive range of sensor modalities to date. Our results establish the scaling laws of LSM for tasks such as imputation, interpolation and extrapolation across both time and sensor modalities. Moreover, we highlight how LSM enables sample-efficient downstream learning for tasks including exercise and activity recognition.

Language models (LMs) are increasingly being deployed to perform autonomous data analyses. However, their data awareness—the ability to recognize, reason over, and appropriately handle data artifacts such as missing values, outliers, and logical inconsistencies—remains underexplored. These artifacts are especially common in real-world tabular data and, if mishandled, can significantly compromise the validity of analytical conclusions. To address this gap, we present RADAR, a benchmark for systematically evaluating data-aware reasoning on tabular data. We develop a framework to simulate data artifacts via programmatic perturbations to enable targeted evaluation of model behavior. RADAR comprises 2,980 table-query pairs, grounded in real-world data spanning 9 domains and 5 data artifact types. In addition to evaluating artifact handling, RADAR systematically varies table size to study how reasoning performance holds when increasing table size. Our evaluation reveals that, despite decent performance on tables without data artifacts, frontier models degrade significantly when data artifacts are introduced, exposing critical gaps in their capacity for robust, data-aware analysis. Designed to be flexible and extensible, RADAR supports diverse perturbation types and controllable table sizes, offering a valuable resource for advancing tabular reasoning.

We present SensorLM, a family of sensor-language foundation models that enable wearable sensor data understanding with natural language. Despite its pervasive nature, aligning and interpreting sensor data with language remains challenging due to the lack of paired, richly annotated sensor-text descriptions in uncurated, real-world wearable data. We introduce a hierarchical caption generation pipeline designed to capture statistical, structural, and semantic information from sensor data. This approach enabled the curation of the largest sensor-language dataset to date, comprising over 59.7 million hours of data from more than 103,000 people. Furthermore, SensorLM extends prominent multimodal pretraining architectures (e.g., CLIP, CoCa) and recovers them as specific variants within a generic architecture. Extensive experiments on real-world tasks in human activity analysis and healthcare verify the superior performance of SensorLM over state-of-the-art in zero-shot recognition, few-shot learning, and cross-modal retrieval. SensorLM also demonstrates intriguing capabilities including scaling behaviors, label efficiency, sensor captioning, and zero-shot generalization to unseen tasks. Code is available at https://github.com/Google-Health/consumer-health-research/tree/main/sensorlm.

We present LuckyChirp, a contactless, passive, opportunistic respiratory tracking solution for commodity device using cascaded sonar modeling. Compared to conventional sonar methods that only solve the respiratory estimation problem (“what is the respiratory rate”), LuckyChirp also solves the additional respiratory detection problem (“is the human present and static enough for respiration sensing”). LuckyChirp uses a custom neural network on pulsed sonar’s wavelet transformed features to detect respiration. The classifier is then cascaded with a respiratory rate estimator. Such holistic design eliminates user friction of manually activating the system and enables passive respiration monitoring for all-day natural use. With Google Nest Hub and Pixel 4 as experimental devices, LuckyChirp achieves a mean absolute error of 0.48±0.98 and 1.07±1.67 breaths/min, respectively, for 20 users participating in a whole-night study. Compared to direct respiratory estimation without respiration classification, this is a ×6 (Nest Hub) and ×4 (Pixel) reduction in error.

Redundant tendon-driven systems such as the human hand or the ACT robotic hand are high-dimensional and nonlinear systems that make traditional control strategies ineffective. The synergy hypothesis from neuroscience suggests that employing dimensionality reduction techniques can simplify the system without a major loss in function. We define a dimensionality reduction framework consisting of separate observation and activation synergies, a first-order model, and an optimal controller. The framework is implemented for two example tasks: adaptive control of thumb posture and hybrid position/force control to enable dynamic handwriting.

We apply path integral reinforcement learning to a biomechanically accurate dynamics model of the index finger and then to the Anatomically Correct Testbed (ACT) robotic hand. We illustrate the applicability of Policy Improvement with Path Integrals (PI2) to parameterized and non-parameterized control policies. This method is based on sampling variations in control, executing them in the real world, and minimizing a cost function on the resulting performance. Iteratively improving the control policy based on real-world performance requires no direct modeling of tendon network nonlinearities and contact transitions, allowing improved task performance.

Muscle synergies are hypothesized as a way to reduce the control space of redundant biological neuromuscular control. Robotics researchers are starting to use grasping synergies to simplify the task of coordinating many joints for complex manipulators. This is especially useful for hand movement control as there are many joints and muscles to control to achieve even simple tasks. This paper uses the Anatomically Correct Testbed (ACT) Robotic Hand to understand how task-specific synergies may be formed and used for complex robotic manipulation tasks such as writing. A comparison is made between synergies formed from (1) general hand movements, (2) task-specific demonstration of generic writing, and (3) practice of writing a specific letter. Results showed that performance using task specific demonstration synergies outperforms performance using general-purpose synergies in terms of completion time, energy expenditure, and trajectory error. Performance using practiced synergy outperforms the performance using demonstration synergies in trajectory error, even though there was no statistical difference in completion time and energy expenditure. These results indicate that task specific synergies from demonstration and practice allow a robotic hand to write better than using more generic synergies that may work for other tasks. How general/specific these synergies should be to optimize the performance of different complex tasks without learning too many specific synergies is an interesting topic for the future.

In the past, the music community conducted research on what makes music more musical or expressive. Much of this work has focused on the manipulation of phrasing, articulation and rubato to make music more expressive. However, it has been difficult to study neuromuscular control used by experts to create such musical music. This paper took a first step toward this effort by using the Anatomically Correct Testbed (ACT) Robotic Hand to mimic the way expert humans play when they are instructed to perform "musically" or "robotically." Results from 22 human subjects showed that musical expression contained a larger range of dynamics and different articulation than robotic expression, while there was no difference in the use of rubato. The ACT Hand was controlled to the level of precision that allowed the replication of expert expressive performance. Its performance was then rated by 17 human listeners against music played by a human expert to show that the ACT Hand could play as musically as an expert human. Furthermore, articulation, phrasing, and rubato were tested in isolation to determine the importance of articulation over phrasing and rubato. This type of study will lead to understanding how to implement future robots to perform musically without pre-programming them, finding ways to teach novice pianists strategies in controlling their muscles to become expressive musicians more quickly, and understanding why humans feel expressiveness or even emotion in music.