we are using a bno055 to measure the rotation of a device around the gravitation axis.
Since there is a lot of metal in the environment and the metal in the environment and other magnetic influences can change, we use the AG (accgyo) mode, without mag.
We have problems with the measuring accuracy, if we rotate the device slowly around the axis, it is very inaccurate and with a rotation of 10 degrees we often have deviations of 2-4 degrees. But if we rotate the device fast, it is very accurate. (Of course, the device was calibrated before, etc.)
Since users are supposed to adjust the angle themselves, it will become difficult to teach the user to rotate the device at a specific speed. In addition, this might seem rather unprofessional and probably leads to considerable inaccuracies.
How can we solve this?
This goal is hard to achieve in fact if move extremely slowly.
The angle rotation was calcuated based on gyro, but if the movement is too slow, then the gyro sensor output is low than certain threshold, we will treat it as noise, so yaw or roll or pitch will keep the same.
The movement deviation will be easily covered by the noise of sensor output. The real movement information you want to find is hard to distingruish here.
So if you need some kind of high accuracy request, you may consider some non consumer sensor.