Robust body position classification during sleep is crucial for closed-loop robotic interventions in position-dependent sleep disorders. This work investigates a compact, custom-made textile pressure sensor to automatically classify recumbent body positions. We implemented a range of traditional classification methods, including Naïve Bayes, Decision Trees, and Support Vector Machines. Furthermore, we trained different machine learning models on recordings from 19 participants, with a varying amount of personalized training data (i.e. from a generalized inter-person classifier towards fully personalized classifiers). We computed the performance metrics F1-score, precision, recall, and accuracy using multi-fold cross-validation for the four classes supine, prone, lateral left, and lateral right, as well as for the binary classes supine vs. non-supine. For the generalized classifier, we could achieve an accuracy of 82.7% for a balanced test set. The personalized models for one male and one female user showed a higher accuracy, namely 95% and 92%. For the binary classifier, the personalized classifiers (male F1-score: 0.97, female F1-score: 0.94) outperformed the generalized classifiers (F1-score: 0.91). Similar to related work, our machine-learning model demonstrated superior performance compared to the three traditional approaches we implemented. Our results show that robust body position classification is possible using small-scale unobtrusive textile bedding sensors, pathing the way for future closed-loop interventions.

Vestibular Stimulation (VS) has been shown to positively affect various autonomic body functions, including sleep. In the past, VS was often investigated using large and complex rocking beds that would only allow for short intervention periods in constrained lab settings. In this work, we present an overview of the mechanics, kinematics, dynamics, and tuning of our latest rocking bed, the Somnomat Casa. Its compact dimensions, comparable to a standard single bed, its connectivity, and easy usability, allow for prolonged studies in home settings investigating the effects of VS during sleep. In a first six-month study with a 12-year-old boy suffering from primary mitochondrial disease with an associated severe sleep disorder, we observed significant improvements in sleep duration (+25 %) and quality of life, as well as a 75 % reduction in caregiving interactions and a 40 % reduction in overall caregiving time. Based on these promising findings, we are currently testing the Somnomat Casa with various patient groups for multiple months each, including elderly with insomnia, children with sleep disorders, Parkinson’s disease patients, and individuals with post-stroke insomnia.

Sleep is essential to boost the rehabilitation outcome as it facilitates motor learning, enhances cognitive performance, and improves mood and well-being. Rocking beds that provide vestibular stimulation may be a promising and non-invasive alternative to conventional pharmaceutical treatments for individuals with sleep problems, offering regenerative sleep without unwanted side effects. Previous research has shown that the effectiveness of the interventions is related to the chosen rocking acceleration. Moreover, the movement of the bed must be comfortable and smooth to avoid disturbing the user's sleep. Previously, the motor control parameters were tuned manually, which was time-consuming, subjective, and did not guarantee minimum deviation from the desired acceleration profile. In this work, we present an efficient and effective method using Gaussian processes to automatically tune the PI control parameters of a rocking bed moving along the longitudinal axis. We first simulated the kinematics of a rocking bed and optimized the control parameters for a chosen objective function that included the desired and the actual accelerations in the movement direction. We then compared the number of iterations needed to reach this objective for a model based on Gaussian processes and for a model based on a naive random exploration of the parameter space. Finally, we implemented the Gaussian process on the rocking bed to automatically tune the control parameters and subjectively compared them to the control parameters that were previously obtained after manual tuning. Our simulation showed that we can reach the control objective after a constant number of iterations using Gaussian processes, independent of the search space size. For the random search, the number of iterations increased quadratically with the size of the search space. The Gaussian process was found to be well transferable to the rocking bed. After less than one hour, control parameters were discovered that outperformed the previous parameters in terms of smoothness. However, despite the smoother motion, the noise emission from the motor, which was not part of the optimization, increased considerably. Our presented technique based on Gaussian processes significantly reduced the time and effort required to optimize the bed's control parameters compared to manual tuning. In future work, the control objective has to be refined to include noise emission as an optimization metric as low noise is an important aspect in sleep-related applications.

Sleep is crucial in rehabilitation processes, promoting neural plasticity and immune functions. Nocturnal body postures can indicate sleep quality and frequent repositioning is required to prevent bedsores for bedridden patients after a stroke or spinal cord injury. Polysomnography (PSG) is considered the gold standard for sleep assessment. Unobtrusive methods for classifying sleep body postures have been presented with similar accuracy to PSG, but most evaluations have been done in research lab environments. To investigate the challenges in the usability of a previously validated device in a clinical setting, we recorded the sleep posture of 17 patients with a sensorized mattress. Ground-truth labels were collected automatically from a PSG device. In addition, we manually labeled the body postures using video data. This allowed us also to evaluate the quality of the PSG labels. We trained neural networks based on the VGG-3 architecture to classify lying postures and used a self-label correction method to account for noisy labels in the training data. The models trained with the video labels achieved a higher classification accuracy than those trained with the PSG labels (0.79 vs. 0.68). The self-label correction could further increase the models' scores based on video and PSG labels to 0.80 and 0.70, respectively. Unobtrusive sensors validated in clinics can, therefore, potentially improve the quality of care for bedridden patients and advance the field of rehabilitation.