[Richard100]

Uutzinger -
Informal testing of BNO055 performance:

Rotational drift ... well ... 0 radians per second
Clamping it to the table resulted in no change in rotation after 10 minutes. By 'no' I mean zero. This is the sensor fusion output (roll, pitch, and yaw, each), not the raw gyro. Gyro itself is about 0.5 deg / min on worst axis, and that's with me doing the rate integration (trapezoidal rule with 20 ms sample period).

Accelerometer repeatability ...
Short term measurement std dev (separating out gravity) is ~ 1 mG for X,Y axes; ~ 4 mG for Z (exposed to Earth G field)
Short term measurement std dev (gravity alone, i.e. separating out linear acceleration) is zero, all axes.
Running sample size of 30.

Log term repeatability (change in mean value after successive power resets) is << Short term SD / doesn't change / not discernible from short term noise.

Magnetometer repeatability ...
I can offer the results of a simple test I conducted with the IMU in full sensor fusion mode (specifically the "NDOF_FMC_OFF" mode, which incorporates the mag with accel and gyro). Strapped the IMU to robot about 12 inches away from CIM drive motors. To generate high current draw, spun robot in place (~500 deg/sec). Resulting Yaw orientation graphed out as flat repeating ramp signal from 0 to 360 deg (0 and 360 are congruent). Implication, if I'm thinking right, is that any robot magnetic field disturbance would cause ramp signal to deviate from flat. The fusion algo appears to automatically discount magnetic disturbance and perhaps fall back to a degraded performance (which I cannot detect).

Strapped IMU directly to CIM motor and repeated test. No change / flat Yaw ramp during spin.

We might be able to come up with a more sophisticated test.

It's entirely possible that the fusion algorithm embedded in the BNO055 outperforms some other algos, don't know. It's a factory's secret sauce in the IMU world. It appears to be very high performing.

At the moment, I'm satisfied with it.