Core storage container for the rsLQR solver. More...

#include <nddata.h>

Data Fields
int	nstates
	size of state vector

int	ninputs
	number of control inputs

int	nsegments
	number of segments, or one less than the length of the horizon

int	depth
	number of columns of factors to store

int	width
	width of each factor. Will be `n` for matrix data and typically 1 for the right-hand-side vector.

double *	data
	pointer to entire chunk of allocated memory

NdFactor *	factors
	(nsegments, depth) array of factors. Stored in column-order.

Detailed Description

Core storage container for the rsLQR solver.

Represents an array of memory blocks, arranged as follows:

\[ \begin{bmatrix} F_1^{(1)} & X_1^{(2)} & \dots & X_1^{(K)} \\ F_2^{(1)} & X_2^{(2)} & \dots & X_2^{(K)} \\ \vdots & \vdots & \ddots & \vdots \\ F_{N-1}^{(1)} & X_{N-1}^{(2)} & \dots & X_{N-1}^{(K)} \\ \end{bmatrix} \]

Each \( F \) is a NdFactor further dividing this memory into chunks of size (n,w) or (m,w), where the width w is equal to NdData.width. Each block is stored as an individual Matrix, which stores the data column-wise. This keeps the data for a single block together in one contiguous block of memory. The entire block of memory for all of the factors is allocated as one large block (with pointer NdData.data).

In the solver, this is used to represent both the KKT matrix data and the right-hand-side vector. When storing the matrix data, each column represents a level of the binary tree. The current implementation only allows for a single right-hand-side vector, so that when width is passed to the initializer, it only creates a single column of factors. Future modifications could alternatively make the right-hand-side vector the last column in the matrix data, as suggested in the original paper.

Methods

The documentation for this struct was generated from the following file:

src/nddata.h

Data Fields

Detailed Description

Methods