Dynamic allocation of Eigen in hard-real time environment

Eigen is a open-source, template-based, c++, matrix library. It has good balance between performance and ease-of-use. We can write matrix formulation intuitively as similar as in Matlab.

Futhermore, Eigen has Geometry module which others don’t have. Especially, I like this feature since I can use quaternion, angle-axis, transformation matrix in robotics application.

However, though above good features, We could get unexpected behavior if we don’t take care of using Eigen in hard real-time environment. One point we have to avoid is a dynamic memory allocation. Since we don’t know how many memory needed when allocated, and it takes time to take heap memory up and down, so it can break real-time loop.

When you declare matrix in Eigen, the below is the simplest way.

1MatrixXf A;

MatrixXf is typedef of Eigen. It is same as

1Matrix<float, Dynamic, Dynamic> A;

The second and thrid argument is row and column size of matrix, respectively. If you declare eigen matrix variable like this, you can resize matrix A as you want. However, the memories for this are dynamically allocated in the backgroud whenever you resize in your application.

Eigen provides fixed-size matrix declaration to avoid this.

1Matrix3f A;

A is 3-by-3 matrix, and is same as

1Matrix<float, 3, 3> A;

I recommend that you should use fixed-size matrix in the hard-real time environment. However, what if you don’t know the exact size of matrix? For this case, Eigen provideto You can use 10-by-10 memory stack and limit the size of matrix by row_size and col_size dynamically. I think stl vector reserve method is similar with this.

In sum, you have to avoid dynamic allocation in the hard real-time environment. To do that, you should carefully use matrix declaration of Eigen library.