Fitted inertia regression line does not match scatter distribution (Roll & Pitch)

[zywzywzyw]

Hi, thanks a lot for this great tool! It really helps to estimate quadrotor parameters from PX4 logs.

When using the tool to estimate inertia for my quadrotor, I noticed that the fitted regression line (red line) does not match the main distribution of the scatter plot (blue points), especially for Roll and Pitch axes.

• For Roll (Ixx):
  The scatter points show a slope with smaller magnitude, but the fitted line is much steeper.
• For Pitch (Iyy):
  A similar mismatch appears: the regression line does not follow the cluster of data points.

Here are two screenshots from my runs (using PX4 ULog, SDLOG_PROFILE = System Identification):

It seems the linear regression might be biased or using outliers in the dataset. Could this be due to:

1. Outlier points with very large predicted torques?
2. Lack of robust regression (RANSAC, weighted LS)?
3. Incorrect normalization of the torque input?

Would it be possible to improve the fitting procedure (e.g., using robust regression or filtering data points) so the red line better represents the main trend of the blue scatter points?

Thanks again for your work on this tool!