dcsam::DCSAM Class Reference

#include <DCSAM.h>

Public Member Functions
	DCSAM ()
	DCSAM (const gtsam::ISAM2Params &isam_params)
void	update (const gtsam::NonlinearFactorGraph &graph, const gtsam::DiscreteFactorGraph &dfg, const DCFactorGraph &dcfg, const gtsam::Values &initialGuessContinuous=gtsam::Values(), const DiscreteValues &initialGuessDiscrete=DiscreteValues())
void	update (const HybridFactorGraph &hfg, const gtsam::Values &initialGuessContinuous=gtsam::Values(), const DiscreteValues &initialGuessDiscrete=DiscreteValues())
void	update (const HybridFactorGraph &hfg, const DiscreteValues &initialGuessDiscrete)
void	update ()
void	updateDiscrete (const gtsam::DiscreteFactorGraph &graph, const gtsam::Values &continuousVals, const DiscreteValues &discreteVals)
void	updateDiscreteInfo (const gtsam::Values &continuousVals, const DiscreteValues &discreteVals)
void	updateContinuous ()
void	updateContinuousInfo (const DiscreteValues &discreteVals, const gtsam::NonlinearFactorGraph &newFactors, const gtsam::Values &initialGuess)
DiscreteValues	solveDiscrete () const
DCValues	calculateEstimate () const
DCMarginals	getMarginals (const gtsam::NonlinearFactorGraph &graph, const gtsam::Values &continuousEst, const gtsam::DiscreteFactorGraph &dfg)
gtsam::DiscreteFactorGraph	getDiscreteFactorGraph () const
gtsam::NonlinearFactorGraph	getNonlinearFactorGraph () const

Constructor & Destructor Documentation

DCSAM() [1/2]

dcsam::DCSAM::DCSAM

(

)

DCSAM() [2/2]

dcsam::DCSAM::DCSAM ( const gtsam::ISAM2Params &  isam_params )

explicit

Member Function Documentation

calculateEstimate()

DCValues dcsam::DCSAM::calculateEstimate

(

)

const

This is the primary function used to extract an estimate from the solver. Internally, calls isam_.calculateEstimate() and dfg_.optimize() to obtain an estimate for the continuous (resp. discrete) variables and packages them into a DCValues pair as (continuousVals, discreteVals).

Returns

a DCValues object containing an estimate of the most probable assignment to the continuous (DCValues.continuous) and discrete (DCValues.discrete) variables.

getDiscreteFactorGraph()

gtsam::DiscreteFactorGraph dcsam::DCSAM::getDiscreteFactorGraph ( ) const

inline

getMarginals()

DCMarginals dcsam::DCSAM::getMarginals	(	const gtsam::NonlinearFactorGraph &	graph,
		const gtsam::Values &	continuousEst,
		const gtsam::DiscreteFactorGraph &	dfg
	)

Used to obtain the marginals from the solver.

NOTE: not obviously correct (see DCSAM.cpp implementation) at the moment. Should perhaps retrieve the marginals for the factor graph obtained as isam_.getFactorsUnsafe() and dfg_ rather than taking as a parameter? I think this was originally intended to mimic the gtsam Marginals class.

Parameters

graph
continuousEst
dfg

getNonlinearFactorGraph()

gtsam::NonlinearFactorGraph dcsam::DCSAM::getNonlinearFactorGraph ( ) const

inline

solveDiscrete()

DiscreteValues dcsam::DCSAM::solveDiscrete

(

)

const

Solve for discrete variables given continuous variables. Internally, calls dfg_.optimize()

Returns

an assignment (DiscreteValues) to the discrete variables in the graph.

update() [1/4]

void dcsam::DCSAM::update

(

)

Simply used to call update without any new factors. Runs an iteration of optimization.

update() [2/4]

void dcsam::DCSAM::update	(	const gtsam::NonlinearFactorGraph &	graph,
		const gtsam::DiscreteFactorGraph &	dfg,
		const DCFactorGraph &	dcfg,
		const gtsam::Values &	initialGuessContinuous = gtsam::Values(),
		const DiscreteValues &	initialGuessDiscrete = DiscreteValues()
	)

For this solver, runs an iteration of alternating minimization between discrete and continuous variables, adding any user-supplied factors (with initial guess) first.

1. Adds new discrete factors (if any) as supplied by a user to the discrete factor graph, then adds any discrete-continuous factors to the discrete factor graph, appropriately initializing their continuous variables to those from the last solve and any supplied by the initial guess.
2. Update the solution for the discrete variables.
3. For all new discrete-continuous factors to be passed to the continuous solver, update/set the latest discrete variables (prior to adding).
4. In one step: add new factors, new values, and earmarked old factor keys to iSAM. Specifically, loop over DC factors already in iSAM, updating their discrete information, then call isam_.update() with the (initialized) new DC factors, any new continuous factors, and the initial guess to be supplied.
5. Calculate the latest continuous variables from iSAM.
6. Update the discrete factors in the discrete factor graph dfg_ with the latest information from the continuous solve.

Parameters

graph	- a gtsam::NonlinearFactorGraph containing any continuous-only factors to add.
dfg	- a gtsam::DiscreteFactorGraph containing any discrete-only factors to add.
dcfg	- a DCFactorGraph containing any joint discrete-continuous factors to add.
initialGuess	- an initial guess for any new continuous keys that. appear in the updated factors (or if one wants to force override previously obtained continuous values).

update() [3/4]

void dcsam::DCSAM::update ( const HybridFactorGraph &  hfg, const DiscreteValues &  initialGuessDiscrete  )

inline

Inline convenience function to allow "skipping" the initial guess for continuous variables while adding an initial guess for discrete variables.

update() [4/4]

void dcsam::DCSAM::update	(	const HybridFactorGraph &	hfg,
		const gtsam::Values &	initialGuessContinuous = gtsam::Values(),
		const DiscreteValues &	initialGuessDiscrete = DiscreteValues()
	)

A HybridFactorGraph is a container holding a NonlinearFactorGraph, a DiscreteFactorGraph, and a DCFactorGraph, so internally this function simply issues a call to update with these internal graphs passed as parameters: that is:

update(hfg.nonlinearGraph(), hfg.discreteGraph(), hfg.dcGraph(), initialGuess);

updateContinuous()

void dcsam::DCSAM::updateContinuous

(

)

At the moment, this just calls isam_.update() internally This could, for example, be used if one wanted to hold the discrete variables constant and perform multiple iterations of updates to the continuous variables.

NOTE: this function behaves quite differently from the similarly named updateDiscrete function, maybe we consider refactoring some of these function names.

updateContinuousInfo()

void dcsam::DCSAM::updateContinuousInfo	(	const DiscreteValues &	discreteVals,
		const gtsam::NonlinearFactorGraph &	newFactors,
		const gtsam::Values &	initialGuess
	)

Given the latest discrete values (dcValues), a set of new factors (newFactors), and an initial guess for any new keys (initialGuess), this function updates the continuous values stored in any DC factors (in the member isam_ instance), marks any affected keys as such, and calls isam_.update with the new factors and initial guess. See implementation for more detail.

NOTE: this is another function that could perhaps be named better.

updateDiscrete()

void dcsam::DCSAM::updateDiscrete	(	const gtsam::DiscreteFactorGraph &	graph = gtsam::DiscreteFactorGraph(),
		const gtsam::Values &	continuousVals = gtsam::Values(),
		const DiscreteValues &	discreteVals = DiscreteValues()
	)

Add factors in graph to member discrete factor graph dfg_, then update any stored continuous variables using those in values by calling updateDiscreteInfo(values).

NOTE: could this be combined with updateDiscreteInfo or do these definitely need to be separate?

Parameters

graph	- a discrete factor graph containing the factors to add
values	- an assignment to the continuous variables (or subset thereof).

updateDiscreteInfo()

void dcsam::DCSAM::updateDiscreteInfo	(	const gtsam::Values &	continuousVals,
		const DiscreteValues &	discreteVals
	)

For any factors in dfg_, update their stored local continuous information with the values from values.

NOTE: could this be combined with updateDiscrete or do these definitely need to be separate?

Parameters

values

- an assignment to the continuous variables (or subset thereof).

---

The documentation for this class was generated from the following files:

• include/dcsam/DCSAM.h
• src/DCSAM.cpp