WIP: Vehicle model estimator (onboard mass & inertia estimation)

[jlecoeur]

(This PR is based on #13219)

It implements the estimation of the vehicle mass and inertia matrix.

To estimate the vehicle inertia, the estimator low-pass filters and computes the time derivative of the vehicle angular acceleration and of the vehicle torque setpoint, then applies a least-mean-square.
Similarly, to estimate the vehicle mass, the estimator low-pass filters and computes the time derivative of the vehicle acceleration and of the vehicle thrust setpoint, then applies a least-mean-square.

This module can be used to estimate the vehicle parameters offline. I would not use its estimate in the controller, at least for now.

FYI @bresch

[mhkabir]

Is it not easier to just weigh a vehicle accurately offline (instead of estimating mass online)?

[jlecoeur]

@mhkabir yes for sure, yet I see a few reasons to do it online:

• if the mixer/allocator is not tuned properly (which is definitely not the case today with the normalized mixers), then vehicle_torque_setpoint and vehicle_thrust_setpoint will not be in the correct units. The mass you measure offline on a scale will not be equal to the mass estimated online.
• inertia is much more complex to measure offline than mass. If we already estimate inertia, it is relatively cheap to also estimate the mass
• if this is reliable enough, you can use the mass estimate in the velocity controller and the inertia estimate in the rate controller, thus have a self-tuning drone
• even if the estimate is not directly fed into the the controllers, you can check its value in a log and refine the tuning manually

[CarlOlsson]

Just curious, if you estimate the inertia based on a torque setpoint and measured angular acceleration, then you assume that the applied torque equals the torque setpoint right? Do you get this from a model of the drivetrain and vehicle dimensions etc?

[jlecoeur]

@CarlOlsson that is the idea yes.
The drivetrain and geometry models exist since #8063. They are used to generate the multirotor mixers. Unfortunately, the actuator_controls did not have units, so the mixers are normalized. I am now trying to get proper units in the torque and thrust setpoints with #13219 and #13351.
Even without proper units the estimation of inertia and mass will be useful to tune the controller. Let's say the normalized torque is in unit U instead of N.m, then the estimated inertia will be expressed in U/(rad.s^-2) instead of N.m/(rad.s^-2) (=kg.m^2). Similarly, let's say the normalized thrust is in unit V instead of N, then the estimated mass will be expressed in V/(m.s^-2) instead of N/(m.s^-2) (=kg).

[jlecoeur]

This PR has been open for a long while without activity. I am closing now.