The clinical requirements of the system were defined based on the interRAI assessment system, a set of comprehensive and standardized geriatric assessment tools for institutional and home settings. RADIO analysed the machine perception state of the art in order to establish which interRAI items can be automatically recognized reliably enough to be useful and what sensor data is needed for these perception technologies. Besides technical feasibility and privacy concerns with respect to the sensor data needed, RADIO carried out a more general discussion on obtrusiveness and how it balances with clinical monitoring requirements.
Clinical and unobtrusiveness requirements guided the design of the RADIO architecture, including the inventory of sensors and perception technologies that RADIO develops and integrates as well as the configurability and modularity of the design. What is stressed in the architecture is the modularity of the design, so that the system can adapt to different environments and to different individual clinical needs and perception of obtrusiveness.
Based on this architectural design, RADIO developed ADL and mood recognition methods that analyse a variety of raw data including audio-visual data, range scans, and text from social media interactions. RADIO also developed and integrated methods for localizing people and objects in the environment based on Bluetooth Low Energy (BLE) beacons, people identification, and home automation device usage. This inventory of elementary methods was fused into high-level recognition systems. The methods are physically instantiated on the robot’s on-board computer, on FPGA implementations, or on off-board Raspberry Pi devices. Together with the home automation sensors, this system of processing units creates a heterogeneous wireless networking environment including ZWave and BLE sensors and WiFi sensors and processors. RADIO investigated both the bridging and the communication robustness of this heterogeneous wireless network.
The RADIO prototype was tested at the Adolfo Muntañá Geriatric Centre, Granollers, Spain and at private homes in Patras, Greece. Regarding usability, both the primary users and caregivers showed a positive attitude towards the RADIO system. Regarding medical efficacy and the reliability of the measurements, the results were also very positive, especially at the more structured piloting room at Granollers. At the Patras pilots, usability from the perspective of the installation and maintenance technician was also tested. The system was proven to be possible to install within one working day in most cases, with few extremely adversarial environments and hardware or network failures hampering installation and operation.
Finally, RADIO proved the concept of connecting instances of the RADIO Home system and medical institutions into the RADIO Ecosystem, adding value to the health data collected by RADIO Homes by making it available not only for medical monitoring but also for medical research. Specifically, RADIO developed network security and access control guidelines for direct access to health data by the competent medical personnel. These were demonstrated through a visualization for monitoring particular primary users’ data, allowing access only to those datapoints for which each caregiver is granted access. Furthermore, RADIO developed the RASSP protocol for privacy-preserving data mining. This protocol allows medical researchers to compute statistics using an R language interface without accessing any of the individual datapoints used in the computation.
