Improving the Reliability of Pervasive Computing Applications By Continuous Checking of Sensor Readings

International audienceThis paper shows that context-aware applications commonly make implicit assumptions about a sensor infrastructure. Because context-awareness critically relies on these assumptions, the developer typically need to ensure their validity by encoding them in the application code, polluting it with non-functional concerns. This defensive programming approach can be avoided by formulating these assumptions aside from the application, thus factorizing them as an explicit model of the sensor infrastructure. This model can be expressed as a set of rules and can be checked automatically and continuously to ensure the reliability of a sensor infrastructure, both at installation time and during normal functioning. The usefulness of our approach is demonstrated in the domain of assisted living for seniors. We applied it to sensor data collected in the context of a 9-month field study of an assisted living platform, deployed at the home of 24 seniors. We show that several kinds of sensor malfunctions could have been identified upon their occurrence, thanks for our continuous checking, and resolved