Now exact derivatives with beautiful functor
parent
3ed5d05b5b
commit
7c2560e977
|
|
@ -116,45 +116,31 @@ pair<Vector3, Vector3> PreintegrationBase::correctMeasurementsBySensorPose(
|
|||
|
||||
//------------------------------------------------------------------------------
|
||||
// See extensive discussion in ImuFactor.lyx
|
||||
Vector9 PreintegrationBase::UpdatePreintegrated(const Vector3& a_body,
|
||||
const Vector3& w_body, double dt,
|
||||
const Vector9& preintegrated,
|
||||
OptionalJacobian<9, 9> A,
|
||||
OptionalJacobian<9, 3> B,
|
||||
OptionalJacobian<9, 3> C) {
|
||||
Vector9 PreintegrationBase::UpdatePreintegrated(
|
||||
const Vector3& a_body, const Vector3& w_body, double dt,
|
||||
const Vector9& preintegrated, OptionalJacobian<9, 9> A,
|
||||
OptionalJacobian<9, 3> B, OptionalJacobian<9, 3> C) {
|
||||
// TODO(frank): expose DexpImpl functor and save on computation
|
||||
static const MultiplyWithInverseFunction<Vector3, 3> applyInvDexp(SO3::ApplyExpmapDerivative);
|
||||
|
||||
const auto theta = preintegrated.segment<3>(0);
|
||||
const auto position = preintegrated.segment<3>(3);
|
||||
const auto velocity = preintegrated.segment<3>(6);
|
||||
|
||||
// Calculate exact mean propagation
|
||||
Matrix3 H;
|
||||
const Matrix3 R = Rot3::Expmap(theta, H).matrix();
|
||||
const Matrix3 invH = H.inverse();
|
||||
Matrix3 H, invH, invHw_H_theta;
|
||||
const Vector invHw = applyInvDexp(theta, w_body, A ? &invHw_H_theta : 0, invH);
|
||||
const Matrix3 R = Rot3::Expmap(theta, A ? &H : 0).matrix();
|
||||
const Vector3 a_nav = R * a_body;
|
||||
const double dt22 = 0.5 * dt * dt;
|
||||
|
||||
Vector9 preintegratedPlus;
|
||||
preintegratedPlus << //
|
||||
theta + invH* w_body* dt, // theta
|
||||
preintegratedPlus << //
|
||||
theta + invHw* dt, // theta
|
||||
position + velocity* dt + a_nav* dt22, // position
|
||||
velocity + a_nav* dt; // velocity
|
||||
|
||||
if (A) {
|
||||
#define USE_NUMERICAL_DERIVATIVE
|
||||
#ifdef USE_NUMERICAL_DERIVATIVE
|
||||
// The use of this yields much more accurate derivatives, but it's slow!
|
||||
// TODO(frank): find a cheap closed form solution (look at Iserles)
|
||||
auto f = [w_body](const Vector3& theta) {
|
||||
return Rot3::ExpmapDerivative(theta).inverse() * w_body;
|
||||
};
|
||||
const Matrix3 invHw_H_theta =
|
||||
numericalDerivative11<Vector3, Vector3>(f, theta);
|
||||
#else
|
||||
// First order (small angle) approximation of derivative of invH*w:
|
||||
// NOTE(frank): Rot3::ExpmapDerivative(w_body) is a less accurate approximation
|
||||
const Matrix3 invHw_H_theta = skewSymmetric(-0.5 * w_body);
|
||||
#endif
|
||||
|
||||
// Exact derivative of R*a with respect to theta:
|
||||
const Matrix3 a_nav_H_theta = R * skewSymmetric(-a_body) * H;
|
||||
|
||||
|
|
|
|||
Loading…
Reference in New Issue