For your question specifically, it would be great to know which example you based this code on, as some of the variables you list are undefined in the code snippet, i.e. “SurfGPS” and “SurfGyro”. Also, I see you are using the “insfilter”, but you may have more success using the “insEKF” filter. See this documentation page for more information, especially the example toward the bottom.
Unfortunately, I don’t think I have enough information to fully answer your question as posed, but I think the information below can be helpful for working with estimation filters in general.
One approach to improving your estimation filter is to compare the filter's estimated state to a truth dataset. A truth dataset might mean simulated data, data collected with your sensors for a pre-determined motion, or data collected with a higher fidelity sensor system. By leveraging a truth dataset as a ground truth for tuning your filter, you can improve your confidence in the filter before using it on field data.
This Detect Multipath GPS Reading Errors Using Residual Filtering in Inertial Sensor Fusion example shows an approach for improving your filter settings by comparing to truth data for GPS and IMU readings.
For data collected using a cell phone, this Estimate Phone Orientation Using Sensor Fusion example walks through some steps for tuning the filter for GPS data from a cell phone. But, this is only using data for orientation-only estimation. This example leverages the “tune” function to tune the filter to the truth data.
To your question about Accelerometer and Gyro biases, these are based on the sensors themselves. They can be determined from the sensor data sheet or with a sensor calibration experiment.
I hope this helps!
