Debuted at CES 2025 in Las Vegas, the Revol smart crib performs several functions such as monitoring vital signs, identifying if the baby’s face is covered, identifying the need for a nappy change, analysing sleep and recognising cries, and using its motorised drive to rock a baby to sleep. Parents are informed of any anomalies via an alert on a smartphone app.
Due for launch this year, the Revol smart crib can also monitor a baby’s longer-term health trends.
Radar is used to capture the vital signs such as heart rate and breath rate. A Bosch spokesperson explained that millimetre-wave radar collects heart rate data by emitting high-frequency electromagnetic waves and analysing the reflections from a baby’s chest and face.
As the chest moves with each heartbeat, these subtle movements alter the reflected signal, allowing the radar to detect changes in distance and motion.
By processing these variations, the radar can accurately measure the heart rate in real-time without physical contact.
The spokesperson added that for cry recognition, the audio signal of an infant crying is different from other voices, such as speech. An AI model was trained, delivering around 86 per cent accuracy based on Bosch’s test cases.
For the nappy change detection scenario Bosch used its indoor air quality BME690 sensor, which detects Volatile Organic Compounds (VOC). It senses the smell of what’s in the baby’s nappy and is said to be more sensitive than a human nose.
For pose detection, Revol detects the position of the arms and legs of the baby. Bosch trained an AI model to ‘read’ that, and the current accuracy is around 85 per cent based on test cases.
Algorithm development
The Revol crib uses algorithms that are categorised into audio-based via microphone and video-based via camera.
In the development phase, Bosch used training data from AudioSet, an open-source dataset with 5,800 hours of audio, including 6.6 hours of infants crying. The testing data included open-source data from Baidu PaddleSpeech, and self-collected infant voices from early adopter users who approved the data collection.
The spokesperson added that BlazeFace and COCO (Common Objects in Context), were used for the video-based algorithm and that the testing data included two parts with a video of a baby toy and self-collected infant video from the early adopters.
“At the start of production, when the algorithm is released and deployed into the Revol Crib, no further training data is needed,” said the spokesperson. “The Crib can predict the results without storing new data.”
Bosch added that all data transmitted is encrypted end-to-end and stored on Bosch-administered servers, while all data at rest is secured locally with individual data encryption keys. The Revol also has an offline mode.
The CTA, organiser of CES 2025, awarded the Revol crib with a CES Innovation Award Honoree in recognition of its design and engineering.
Comment: Autonomous construction requires open data standards
Autonomous construction works best when there's open data standards so different systems can work together seamlessly. Without it, machines and...