Detailed Description

This class implements a Rao-Blackwelized Particle Filter (RBPF) approach to map building (SLAM).

Internally, the list of particles, each containing a hypothesis for the robot path plus its associated metric map, is stored in an object of class CMultiMetricMapPDF.

This class processes robot actions and observations sequentially (through the method CMetricMapBuilderRBPF::processActionObservation) and exploits the generic design of metric map classes in MRPT to deal with any number and combination of maps simultaneously: the likelihood of observations is the product of the likelihood in the different maps, etc.

A number of particle filter methods are implemented as well, by selecting the appropriate values in TConstructionOptions::PF_options. Not all the PF algorithms are implemented for all kinds of maps.

For an example of usage, check the application "rbpf-slam", in "apps/RBPF-SLAM". See also the wiki page.

Note:: Since MRPT 0.7.2, the new variables "localizeLinDistance,localizeAngDistance" are introduced to provide a way to update the robot pose at a different rate than the map is updated.; Since MRPT 0.7.1 the semantics of the parameters "insertionLinDistance" and "insertionAngDistance" changes: the entire RBFP is now NOT updated unless odometry increments surpass the threshold (previously, only the map was NOT updated). This is done to gain efficiency.; Since MRPT 0.6.2 this class implements full 6D SLAM. Previous versions worked in 2D + heading only.

See also:: CMetricMap

Definition at line 65 of file CMetricMapBuilderRBPF.h.

#include <mrpt/slam/CMetricMapBuilderRBPF.h>

Inheritance diagram for mrpt::slam::CMetricMapBuilderRBPF:

[legend]

List of all members.

Classes
struct	TConstructionOptions
	Options for building a CMetricMapBuilderRBPF object, passed to the constructor. More...
struct	TStats
	This structure will hold stats after each execution of processActionObservation. More...
Public Member Functions
	CMetricMapBuilderRBPF (const TConstructionOptions &initializationOptions)
	Constructor.
virtual	~CMetricMapBuilderRBPF ()
	Destructor.
void	initialize (const CSimpleMap &initialMap=CSimpleMap(), CPosePDF *x0=NULL)
	Initialize the method, starting with a known location PDF "x0"(if supplied, set to NULL to left unmodified) and a given fixed, past map.
void	clear ()
	Clear all elements of the maps.
CPose3DPDFPtr	getCurrentPoseEstimation () const
	Returns a copy of the current best pose estimation as a pose PDF.
void	getCurrentMostLikelyPath (std::deque< TPose3D > &outPath) const
	Returns the current most-likely path estimation (the path associated to the most likely particle).
void	processActionObservation (CActionCollection &action, CSensoryFrame &observations)
	Appends a new action and observations to update this map: See the description of the class at the top of this page to see a more complete description.
void	getCurrentlyBuiltMap (CSimpleMap &out_map) const
	Fills "out_map" with the set of "poses"-"sensory-frames", thus the so far built map.
CMultiMetricMap *	getCurrentlyBuiltMetricMap ()
	Returns the map built so far.
unsigned int	getCurrentlyBuiltMapSize ()
	Returns just how many sensory-frames are stored in the currently build map.
void	saveCurrentEstimationToImage (const std::string &file, bool formatEMF_BMP=true)
	A useful method for debugging: the current map (and/or poses) estimation is dumped to an image file.
void	drawCurrentEstimationToImage (utils::CCanvas *img)
	A usefull method for debugging: draws the current map and path hypotheses to a CCanvas.
void	saveCurrentPathEstimationToTextFile (std::string fil)
	A logging utility: saves the current path estimation for each particle in a text file (a row per particle, each 3-column-entry is a set [x,y,phi], respectively).
double	getCurrentJointEntropy ()
Public Attributes
CMultiMetricMapPDF	mapPDF
	The map PDF: It includes a path and associated map for each particle.
TStats	m_statsLastIteration
	This structure will hold stats after each execution of processActionObservation.
Protected Attributes
bayes::CParticleFilter::TParticleFilterOptions	m_PF_options
	The configuration of the particle filter:
float	insertionLinDistance
	Distances (linear and angular) for inserting a new observation into the map.
float	insertionAngDistance
float	localizeLinDistance
	Distances (linear and angular) for updating the robot pose estimate (and particles weighs, if applicable).
float	localizeAngDistance
mrpt::poses::CPose3DPDFGaussian	odoIncrementSinceLastLocalization
	Traveled distance since last localization update.
mrpt::poses::CPose3D	odoIncrementSinceLastMapUpdate
	Traveled distance since last map update.
non_copiable_ptr< CMultiMetricMap >	currentMetricMapEstimation
	A buffer: memory is actually hold within "mapPDF".

Constructor & Destructor Documentation

mrpt::slam::CMetricMapBuilderRBPF::CMetricMapBuilderRBPF ( const TConstructionOptions & initializationOptions )

Constructor.

virtual mrpt::slam::CMetricMapBuilderRBPF::~CMetricMapBuilderRBPF ( ) [virtual]

Destructor.

Member Function Documentation

void mrpt::slam::CMetricMapBuilderRBPF::clear ( )

Clear all elements of the maps.

Reimplemented from mrpt::slam::CMetricMapBuilder.

void mrpt::slam::CMetricMapBuilderRBPF::drawCurrentEstimationToImage ( utils::CCanvas * img )

A usefull method for debugging: draws the current map and path hypotheses to a CCanvas.

double mrpt::slam::CMetricMapBuilderRBPF::getCurrentJointEntropy ( )

void mrpt::slam::CMetricMapBuilderRBPF::getCurrentlyBuiltMap ( CSimpleMap & out_map ) const [virtual]

Fills "out_map" with the set of "poses"-"sensory-frames", thus the so far built map.

Implements mrpt::slam::CMetricMapBuilder.

unsigned int mrpt::slam::CMetricMapBuilderRBPF::getCurrentlyBuiltMapSize ( ) [virtual]

Returns just how many sensory-frames are stored in the currently build map.

Implements mrpt::slam::CMetricMapBuilder.

CMultiMetricMap* mrpt::slam::CMetricMapBuilderRBPF::getCurrentlyBuiltMetricMap ( ) [virtual]

Returns the map built so far.

NOTE that for efficiency a pointer to the internal object is passed, DO NOT delete nor modify the object in any way, if desired, make a copy of ir with "duplicate()".

Implements mrpt::slam::CMetricMapBuilder.

void mrpt::slam::CMetricMapBuilderRBPF::getCurrentMostLikelyPath ( std::deque< TPose3D > & outPath ) const

Returns the current most-likely path estimation (the path associated to the most likely particle).

CPose3DPDFPtr mrpt::slam::CMetricMapBuilderRBPF::getCurrentPoseEstimation ( ) const [virtual]

Returns a copy of the current best pose estimation as a pose PDF.

Implements mrpt::slam::CMetricMapBuilder.

void mrpt::slam::CMetricMapBuilderRBPF::initialize	(	const CSimpleMap &	initialMap = `CSimpleMap()`,
		CPosePDF *	x0 = `NULL`
	)		`[virtual]`

Initialize the method, starting with a known location PDF "x0"(if supplied, set to NULL to left unmodified) and a given fixed, past map.

Implements mrpt::slam::CMetricMapBuilder.

void mrpt::slam::CMetricMapBuilderRBPF::processActionObservation	(	CActionCollection &	action,
		CSensoryFrame &	observations
	)		`[virtual]`

Appends a new action and observations to update this map: See the description of the class at the top of this page to see a more complete description.

Parameters:

action	The incremental 2D pose change in the robot pose. This value is deterministic.
observations	The set of observations that robot senses at the new pose. Statistics will be saved to statsLastIteration

Implements mrpt::slam::CMetricMapBuilder.

void mrpt::slam::CMetricMapBuilderRBPF::saveCurrentEstimationToImage	(	const std::string &	file,
		bool	formatEMF_BMP = `true`
	)		`[virtual]`

A useful method for debugging: the current map (and/or poses) estimation is dumped to an image file.

Parameters:

file	The output file name
formatEMF_BMP	Output format = true:EMF, false:BMP

Implements mrpt::slam::CMetricMapBuilder.

void mrpt::slam::CMetricMapBuilderRBPF::saveCurrentPathEstimationToTextFile ( std::string fil )

A logging utility: saves the current path estimation for each particle in a text file (a row per particle, each 3-column-entry is a set [x,y,phi], respectively).