The Living Somnox

 1. Introduction

Insomnia has many consequences for society. Scientific research shows that the consequences of a bad night’s sleep are noticeable in, for example, the productivity of employees, road safety, the effect of therapies, the use of medication and the performance of students during their education and in the quality of life in in general. In times of crisis, high pressure on the labor market, Covid-19 and other stressful situations, the vitality of the population is of great importance for the resilience and recovery capacity of society as a whole. This importance applies to medical health, psychological health as well as to the joy of life of individuals and thus the socializing power of citizens.
After the successful market introduction of the Somnox 1 and Somnox 2, Somnox has decided in 2023 to proceed with a structural and innovative redevelopment of the sleep robot into an ‘intelligent sleep robot’. This so-called ‘Living Somnox’ will use IoT, sensor technology and AI. Through the collaboration with Annovating and the expertise that becomes available, the latest challenges can be solved and the Somnox can take a big step forward. Because the Living Somnox also contains intelligent software after completion of this project, the effectiveness of the robot will increase and with it customer satisfaction. With this, the Netherlands is developing one of the first self-learning physical sleep robots, which means that the current Dutch competitive position for sleep robots (other than just an App) can be strengthened.
Insomnia is a very big problem worldwide. In the Netherlands, 63% of the population is not satisfied with the quality of sleep and 1 in 4 inhabitants suffers from some form of insomnia (CBS, 2009). In the lower income groups, the percentage of sleep problems is even increasing (CBS, 2018). This problem is not only limited to the Netherlands. In Sweden, for example, about 41% (4,000,000 people) of the population also suffer from insomnia, compared to 33% (26,000,000 people) in Germany, 34% (22,000,000 people) in France and 36% (23,000,000 people) in the United Kingdom. So it is a huge problem all over Western Europe. In the United States, it’s an even bigger problem; 56% of the inhabitants (178,000,000 people) suffer from some form of insomnia. It can therefore be said that insomnia is a worldwide problem (Léger, 2008). A lack of sleep can lead to little energy during the day, but in serious cases also to depression, reduced work productivity and traffic and industrial accidents. More than 2 million people in the Netherlands try to get a better night’s sleep with medication (CBS, 2009). However, these drugs can be addictive and can have unpleasant side effects. A better solution needs to be found for this. Often people are very busy during the day with work and their social life. At night they should relax and rest, but often this does not happen. Due to stress or anxiety, people have trouble falling and staying asleep. Their heart rate goes up and they start breathing faster. And when you can’t sleep, it stresses you out again and a vicious cycle begins. It is precisely this vicious circle that we break with the new Living Somnox.

2. Project

In this project, Somnox and Annovating B.V are working together on the development of the Living Somnox. Annovating takes care of the AI and data science part of the development. For this purpose, the database is linked to the Somnox server to exchange data. Patients with a Somnox will also wear a fitness tracker. The data from selfb is processed in Annovating’s AI software called ‘LorAine’. The results from LorAIne are fed back to the Somnox server and ultimately to the individual sleep robots.

3. Main question

Can applying ML to heart rate data from fitness trackers combined with accelerometer and gyroscope data from a Somnox sleep robot improve the performance of a Somnox sleep robot?


  1. What data preprocessing should be applied?
  2. Is it preferable to work with features or is it better to let a neural network learn the raw data?
  3. Does a Classifier provide better results for this question than a Recurrent Neural Network?
  4. Is it possible to quantify with a Python model to what extent the Main Question can be answered positively?

Other potential sub-questions

  1. HR data and accelerometer data are measured at different frequencies. Does resampling need to be done? With what frequency?
  2. Before we apply ML, it is possible to normalize per data subject over the ‘days’. Wouldn’t it be better to normalize vertically per ‘feature’?
  3. Are there any ‘outliers’ in terms of data subjects? In other words, if there are data subjects that differ greatly from the others, they may disrupt ‘the ML model’. Comparable to learning Chinese from a bad Chinese textbook. Could you identify those outliers and improve the model (sensitivity and selectivity) by eliminating them? How could you automate that process?

5. Performance

  1. What is the performance of the ML models used?
  2. Choose two algorithms and apply hyperparameter tuning. Would you use grid search for this or better (and faster?) random search? Why?
  3. Apply a PCA and rank the ‘value’ of the features used.
  4. Develop a module that produces AOCROC calculations and figures.

6. Annovating B.V

Annovating B.V ( has existed since 2017 and was founded by Ir. Auke de Leeuw. Mr. de Leeuw has an academic background in the aerospace industry and has been active in company management and commercial technical positions for more than 20 years. Since 2017, Mr De Leeuw has been involved with Annovating in the development of medical ICT and artificial intelligence. Annovating B.V develops and manages the data platform and now also facilitates research in the Hagaziekenhuis and at TU Delft. The core activity at the moment is the development of software to detect atrial fibrillation and drug-unsafe situations from data from fitness trackers. Artificial intelligence is also used for this by Annovating. As a business partner at Annovating, Mr. Taco Kind, doctor and electrical engineer graduated from TU Delft, is also involved. Annovating also works together with MvAutomatisering. MVAutomatisering takes care of the management of the database and all server side applications of Annovating to keep the existing services operational.

Data Scientist
Hani Al-Ers is a researcher in the field of human-machine interactions. He completed his PhD at the Delft University of Technology at the Interactive Intelligence group of the Faculty Computer Science (EEMCS). Philips Research in Eindhoven sponsored his project which was aimed at improving the user experience of Philips tv sets. He completed 2 post-docs at the Delft University of Technology, during which he managed international consortia on topics such as an improved quality of life for the elderly. Currently, he is conducting research in the field of health and education and he leads the Research Education activities at the Dutch Innovation Factory.