Advanced Debugging Architecture for Smart Inertial Sensors using Sensor-in-the-Loop

Smart inertial sensors have emerged in the last years enabling developers to implement sensor fusion or tasks like gesture detection directly in the sensor hardware and thus reducing the processing latency and energy consumption. The development of software for smart inertial sensors however faces difficulties from the limited hardware capabilities of the mu C hardware and the lack of options to conduct reproducible tests directly on the hardware. We propose a Sensor-in-the-Loop architecture that allows a developer to record data from a smart sensor and inject the recorded data back into the sensor at a later time to evaluate the performance of the software in a reproducible way. With the proposed architecture it is possible to examine the performance and computational load on real hardware with recorded data thus allowing developers to optimize and improve the software in a more targeted way. In our implementation, the memory overhead for the code instrumentalization is just 0.63 %. The used RTT interface is at least four times faster than the regular SPI sensor interface. In experiments, we show that our proposed architecture is able to record and inject data of up to three 3-DOF sensors with 1.6 kHz sampling frequency in real time.