Inconsistencies between crba() and rena()

[Mr-Red331]

Hi, This is a question about the inconsistencies between crba() and rena(). I used two ways (rnea and crba) to calculate tau, but the result was inconsistent:

1. tau = rnea(model, data, q, v, a, extf)
2. tau = Ma + Cv + G - JT*extf

Here is the specific process:

1. define a set a const random input data: q, v, a, extf;
2. calculate matrix M with crba(model, data, q);
3. calculate matrix C with computeCoriolisMatrix(model, data, q, v,);
4. calculate matrix G with computeGeneralizedGravity(model, data, q);
5. calculate matrix J with computeJointJacobian(model, data, q, joint, J) and getJointJacobian(model, data, joint, LOCAL, J);
6. calculate tau_crba = M*a + C*v + G - JT*extf;
7. calculate tau_rnea = rnea(model, data, q, v, a, extf);

But, the final result tau_crba != tau_rnea.

At the same time, there is an interesting phenomenon: I let a.setZero() and the rest q, v, extf remains const random. After going through the above steps, I could get the result tau_crba = tau_rnea.

So, is there any wrong in calculating M or define a?

[jcarpent]

In C++, only the upper part of the mass matrix is filled, for performance issues. You can use the Eigen API to consider the full matrix (M.selfadjointView<Eigen::Upper>())

[Mr-Red331]

Thanks! This method worked and my problem was solved.

And futhermore, I would be very grateful if you could tell me why this method could solve this problem.

[jcarpent]

Just display the mass matrix, and you will see that the lower part is full of 0s.

[Mr-Red331]

That means, in C++, the matrix M calculated with crba() is unfilled and incorrect, and only the filled Matrix M is the correct for calculate dynamics?

[fabinsch]

This behavior is expected and described here in the official pinocchio documentation.

[Mr-Red331]

Thanks very much, that is very helpful to me!