#include <node.hpp>
Public Types | |
| enum | nodeMode { STANDARD = 0, ANCHOR, SCHUR_MARGI } |
Public Member Functions | |
| Node (uint_t dim, nodeMode mode=STANDARD) | |
| virtual void | update (VectRefConst &dx)=0 |
| virtual void | update_from_auxiliary (VectRefConst &dx)=0 |
| virtual void | set_state (MatRefConst &x)=0 |
| virtual void | set_auxiliary_state (MatRefConst &x)=0 |
| virtual const MatRefConst | get_state () const =0 |
| virtual MatRefConst | get_auxiliary_state () const =0 |
| virtual void | print () const |
| factor_id_t | get_id () const |
| void | set_id (factor_id_t id) |
| factor_id_t | get_dim (void) const |
| bool | is_connected_to_EF () const |
| void | set_connected_to_EF (bool state=true) |
| void | set_node_mode (nodeMode mode) |
| nodeMode | get_node_mode () |
Protected Attributes | |
| factor_id_t | id_ |
| uint_t | dim_ |
| nodeMode | node_mode_ |
| bool | isConnected2EF_ |
Detailed Description
Node class is an abstract class for creating future nodes. Pure abstract methods on get and set dimension is just a reminder that this is an Abstract class
Node does not track the factors associated to it, it just contains the state variables and allows to update its value. (whereas before we kept track of factors in an unnecessary complex solution).
States are indicated as references to MatX matrices, since it can be either a block vector or a matrix (transformation matrix)
For updates it will always be a block vector : Ref < MatX1 >
Two states are kept at the same time:
- (principal) state: used for factors evaluations, errors and Jacobians
- auxiliary state: a book-keep state useful for partial updates
Node mode refer on how they will be processed further in the FGraph:
- Standard: process as usual
- Anchor: This node will be constant and insensitive to gradients (not processed). It must be correctly initialized
- Schur_margi: Node to be marginalized when using Schur
Member Function Documentation
◆ get_auxiliary_state()
|
pure virtual |
Returns a matrix to the last auxiliary state. This data structure is for the incre- metal implementation, or for error evaluation
Implemented in mrob::NodePose3d, mrob::NodePlane4d, mrob::NodeLandmark2d, mrob::NodePose2d, and mrob::NodeLandmark3d.
◆ get_state()
|
pure virtual |
Declared as a dynamic matrix reference to allow any size to be returned. At run time returns a Reference to a fixed size matrix and provide it as an argument for the getState function, no need to be dynamic, as long as the dimension is correctly set
Implemented in mrob::NodePose3d, mrob::NodePlane4d, mrob::NodeLandmark2d, mrob::NodePose2d, and mrob::NodeLandmark3d.
◆ is_connected_to_EF()
|
inline |
Methods for interacting with Eigen factors, a special kind of factors. This bool variable and its associated methods will make easier the construction of the matrix TODO not used right now, but it could to improve efficiency.
◆ set_auxiliary_state()
|
pure virtual |
New auxiliary state set
Implemented in mrob::NodePose3d, mrob::NodePlane4d, mrob::NodeLandmark2d, mrob::NodePose2d, and mrob::NodeLandmark3d.
◆ set_state()
|
pure virtual |
At run time sets the new value of the estate and auxiliary to be x
Implemented in mrob::NodePose3d, mrob::NodePlane4d, mrob::NodeLandmark2d, mrob::NodePose2d, and mrob::NodeLandmark3d.
◆ update()
|
pure virtual |
The update function, given any block vector it updates the value of the state (principal).
Since we don't know the size at compilation, we declare a dynamic matrix, but on run-time we would like to use a fixed block matrix, and this virtual function will handle it nicely.
Implemented in mrob::NodePose3d, mrob::NodePlane4d, mrob::NodeLandmark2d, mrob::NodePose2d, and mrob::NodeLandmark3d.
◆ update_from_auxiliary()
|
pure virtual |
Updates FROM the auxiliary state the principal state.
Implemented in mrob::NodePose3d, mrob::NodePlane4d, mrob::NodeLandmark2d, mrob::NodePose2d, and mrob::NodeLandmark3d.
Member Data Documentation
◆ isConnected2EF_
|
protected |
On this pure abstract class we can't define a vector state, but we will return and process Ref<> to dynamic matrices. The reason for doing that is polymorphism. We prefer to preserve polymorphism for node storage at the cost of a returning Ref, although there is no extra allocation on this process. For instance, we will declare: Mat61 x_; or SE3 T_; (for transformations)
The documentation for this class was generated from the following files:
- src/FGraph/mrob/node.hpp
- src/FGraph/node.cpp
