There seems to be a mistake in how the derivative of the normal/tangent is computed with respect to the x or y directions. The derivative of the normalized vector (n_norm) wrt the non-normalized one (n) is a 3 x 3 matrix that should be fully multiplied with the vector (dn_x/dx, dn_y/dx, dn_z/dx) while currently only a single column is multiplied elementwise. This is only an issue for terrains with non-zero second-order derivatives. Changing these lines resulted in fewer incorrect gradient values according to IPOpt's gradient checker for me.
Thanks for writing such a nice piece of software!
There seems to be a mistake in how the derivative of the normal/tangent is computed with respect to the x or y directions. The derivative of the normalized vector (n_norm) wrt the non-normalized one (n) is a 3 x 3 matrix that should be fully multiplied with the vector (dn_x/dx, dn_y/dx, dn_z/dx) while currently only a single column is multiplied elementwise. This is only an issue for terrains with non-zero second-order derivatives. Changing these lines resulted in fewer incorrect gradient values according to IPOpt's gradient checker for me.