I’m using Torque for some driving experiments and looking for trends in acceleration. My focus at the moment is to figure out the best representation of device acceleration, using OBD port acceleration as a “ground truth.” We used an Android Nexus 5X device mounted to the side of the car’s passenger seat window tilted slightly forward. All data was collected by driving in a straight line in a parking lot. At the beginning of the drive we used the built-in calibration feature.

The car being used is a Toyota Prius. We have G(x), G(y), and G(z), along with Acceleration Sensor(X axis)(g), Acceleration Sensor(Y axis)(g), and Acceleration Sensor(Z axis)(g). Plotting the G acceleration and the Acceleration Sensor measurements next to each other indicates that they’re measuring the same thing minus some drift over time. The black curve is the OBD port acceleration.

Looking at this plot, it appears that calibration sets the x-axis to vehicle acceleration, y-axis to gravity, and z-axis to 0, but seems to loosely follow the same trend as the x-direction. However taking the dot product of x and z results in a worse fit than using just x

Does Torque calibrate acceleration by setting the x-direction to acceleration, y to gravity, and z to zero? And if not, how does Torque calibrate acceleration?