RBPF-SLAM算法(C ++库mrpt-slam)

  • RBPF-SLAM算法(C ++库mrpt-slam)已关闭评论
  • 3,091 views
  • A+
所属分类:工业4.0

1.数学背景
阅读资料:

“动态贝叶斯网络的Rao-Blackwellised粒子滤波”,Doucet,A。和De Freitas,N。和Murphy,K。和Russell,S。(2000) - PDF
Boris Lipchin的一些幻灯片。

2.在MRPT中实施基于RBPF的SLAM解决方案
RBPF是粒子滤波器的一个特殊实例,因此原则 上可以使用虚拟C ++基类中 声明的四种通用滤波算法中的 任何一种来顺序估计机器人路径。只是对算法的简要总结:

序贯重要性重采样 - SIR与“标准”的建议分布(枚举值: pfStandardProposal)
辅助粒子滤波 -与“标准”建议分布APF(枚举值: pfAuxiliaryPFStandard)
最佳建议分布 (枚举值: pfOptimalProposal)
近似最优采样 - A-基于采样排斥逼近最佳建议时的闭合形式的表达不可用(枚举值: pfAuxiliaryPFOptimal)
注意:到目前为止(2011年8月),APF算法没有实现RBPF-SLAM(这反过来对Monte Carlo本地化有用 )。

2.1。 pfStandardProposal:带有“标准提案”的SIR和任何类型的指标图
描述: 只要在每张地图上为传感器观测实现观测似然函数, 该 pfStandardProposal 算法可以与任何度量地图或其中几个地图的组合一起使用 。然而,这种算法并不值得推荐,因为“标准”提案分布是运动模型(“ 动作 ”),并且大多数粒子将以与观测不相容的区域结束,因此重采样将经常发生。只有在已知运动模型非常精确的情况下才使用该算法。 配置文件示例:

2.2。 pfAuxiliaryPFOptimal:使用任何类型的公制映射的近似最优建议
说明: 只要在每张地图中为传感器观测实现观测似然函数 , pfAuxiliaryPFOptimal 算法(本文 和 本文中介绍的 )可以与任何度量地图或其中几个地图的组合一起使用。无论运动模型的质量如何,算法都会自动逼近最优投标分布。然而,运动模型越差,运行每次迭代所需的时间就越长。 配置文件示例:

#------------------------------------------------------------
# Config file for the "RBPF-SLAM" application
# See: http://www.mrpt.org/list-of-mrpt-apps/application-rbpf-slam/
#------------------------------------------------------------


#=======================================================
# Section: [MappingApplication]
# Use: Here comes global parameters for the app.
#=======================================================
[MappingApplication]
# The source file (RAW-LOG) with action/observation pairs
rawlog_file=../../datasets/2006-01ENE-21-SENA_Telecom Faculty_one_loop_only.rawlog
rawlog_offset=0

# The directory where the log files will be saved (left in blank if no log is required)
logOutput_dir=LOG_GRIDMAPPING
LOG_FREQUENCY=10			// The frequency of log files generation:
GENERATE_LOG_JOINT_H=0
GENERATE_LOG_INFO=0
SAVE_MAP_IMAGES=1
SAVE_3D_SCENE=1
SAVE_POSE_LOG=0
SAVE_ENOSE_READINGS=0
CAMERA_3DSCENE_FOLLOWS_ROBOT=0
SHOW_PROGRESS_IN_WINDOW=1

insertionLinDistance     = 1.0	// The distance threshold for inserting observations in the map (meters)
insertionAngDistance_deg = 40.0	// The distance threshold for inserting observations in the map (degrees)

localizeLinDistance     = 1.0	// The distance threshold for actually running the PF, otherwise will rely on odometry  (meters)
localizeAngDistance_deg = 40.0	// The rotational threshold for actually running the PF, otherwise will rely on odometry  (degrees)

# Console verbosity level: 
# One of: DEBUG, INFO (default), WARN, ERROR
verbosity_level = DEBUG

#----------------------------------------------------------------------------------
# The Particle Filter algorithm:
#	0: pfStandardProposal
#	1: pfAuxiliaryPFStandard
#	2: pfOptimalProposal      *** (ICP-based (Grisetti's method),...)
#	3: pfAuxiliaryPFOptimal	  *** (Optimal SAMPLING)
#
# See: http://babel.isa.uma.es/mrpt/index.php/Particle_Filter_Algorithms
#----------------------------------------------------------------------------------
PF_algorithm		= 3
adaptiveSampleSize	= 0		// 0: Fixed # of particles, 1: KLD adaptive

#----------------------------------------------------------------------------------
# The Particle Filter Resampling method:
#	0: prMultinomial
#	1: prResidual
#	2: prStratified
#	3: prSystematic
#
# See: http://babel.isa.uma.es/mrpt/index.php/Resampling_Schemes
#----------------------------------------------------------------------------------
resamplingMethod=0

sampleSize=25			// Sample size (for fixed number)
BETA=0.50			// Resampling ESS threshold


# ========================================================
#            MULTIMETRIC MAP CONFIGURATION
# See docs for (Google for) mrpt::maps::CMultiMetricMap
# ========================================================
# Creation of maps:
occupancyGrid_count=1
gasGrid_count=0
landmarksMap_count=0
beaconMap_count=0
pointsMap_count=0

# Selection of map for likelihood: (fuseAll=-1,occGrid=0, points=1,landmarks=2,gasGrid=3)
likelihoodMapSelection=-1


# ====================================================
#         CHybridMetricMapPDF::TPredictionParams
# ====================================================
powFactor=0.01					// A "power factor" for updating weights
pfAuxFilterOptimal_MaximumSearchSamples=400	// For PF algorithm=3


# -----------------------------------------------------------------
#   pfOptimalProposal_mapSelection
#  Only for PF algorithm=2 (Exact "pfOptimalProposal")
#   Select the map on which to calculate the optimal 
#    proposal distribution. Values:
#   0: Gridmap   -> Uses Scan matching-based approximation (based on Stachniss' work)
#   1: Landmarks -> Uses matching to approximate optimal
#   2: Beacons   -> Used for exact optimal proposal in RO-SLAM
# -----------------------------------------------------------------
pfOptimalProposal_mapSelection=0

# Adaptive sample size parameters ------------------
KLD_maxSampleSize=150
KLD_minSampleSize=15
KLD_binSize_XY=0.03
KLD_binSize_PHI_deg=1
KLD_delta=0.001
KLD_epsilon=0.5


# ====================================================
#   MULTIMETRIC MAP: OccGrid #00
# ====================================================
# Creation Options for OccupancyGridMap 00:
[MappingApplication_occupancyGrid_00_creationOpts]
resolution=0.07
disableSaveAs3DObject=0


# Insertion Options for OccupancyGridMap 00:
[MappingApplication_occupancyGrid_00_insertOpts]
mapAltitude					= 0
useMapAltitude				= 0
maxDistanceInsertion		= 35
maxOccupancyUpdateCertainty	= 0.60
considerInvalidRangesAsFreeSpace = 1
minLaserScanNoiseStd		= 0.001
wideningBeamsWithDistance 	= 0


# Likelihood Options for OccupancyGridMap 00:
[MappingApplication_occupancyGrid_00_likelihoodOpts]
likelihoodMethod=4		// 0=MI, 1=Beam Model, 2=RSLC, 3=Cells Difs, 4=LF_Trun, 5=LF_II
LF_decimation=5
LF_stdHit=0.20
LF_maxCorrsDistance=0.30
LF_zHit=0.999
LF_zRandom=0.001
LF_maxRange=30
LF_alternateAverageMethod=0

enableLikelihoodCache=1

使用一台SICK激光扫描仪和一个中型(约100米长)环路封闭的网格映射示例。

#------------------------------------------------------------
# Config file for the "RBPF-SLAM" application
# See: http://www.mrpt.org/list-of-mrpt-apps/application-rbpf-slam/
#------------------------------------------------------------


#=======================================================
# Section: [MappingApplication]
# Use: Here comes global parameters for the app.
#=======================================================
[MappingApplication]
# The source file (RAW-LOG) with action/observation pairs
#rawlog_file=../datasets/2007-05MAY-17_SENA_Floor2.3_3lasers.rawlog
rawlog_file=J:/Trabajo/RAWLogs/2007-05MAY-17_SENA_Floor2.3_3lasers.rawlog
rawlog_offset=0

# The directory where the log files will be saved (left in blank if no log is required)
logOutput_dir=LOG_3GRIDMAPS
LOG_FREQUENCY=5			// The frequency of log files generation:
GENERATE_LOG_JOINT_H=0
GENERATE_LOG_INFO=0
SAVE_MAP_IMAGES=1
SAVE_3D_SCENE=1
SAVE_POSE_LOG=0
SAVE_ENOSE_READINGS=0
CAMERA_3DSCENE_FOLLOWS_ROBOT=0
SHOW_PROGRESS_IN_WINDOW=1

insertionLinDistance=0.1	// The distance threshold for inserting observations in the map (meters)
insertionAngDistance_deg=4.0	// The distance threshold for inserting observations in the map (degrees)

localizeLinDistance     = 0.1	// The distance threshold for actually running the PF, otherwise will rely on odometry  (meters)
localizeAngDistance_deg = 4.0	// The rotational threshold for actually running the PF, otherwise will rely on odometry  (degrees)

# Console verbosity level: 
# One of: DEBUG, INFO (default), WARN, ERROR
verbosity_level = DEBUG

#----------------------------------------------------------------------------------
# The Particle Filter algorithm:
#	0: pfStandardProposal
#	1: pfAuxiliaryPFStandard
#	2: pfOptimalProposal      *** (ICP,...)
#	3: pfAuxiliaryPFOptimal	  *** (Optimal Aux)
#
# See: /docs/html/topic_pf.html  or http://mrpt.sourceforge.net/topic_pf.html
#----------------------------------------------------------------------------------
PF_algorithm=3
adaptiveSampleSize=0		// 0: Fixed # of particles, 1: KLD adaptive

#----------------------------------------------------------------------------------
# The Particle Filter Resampling method:
#	0: prMultinomial
#	1: prResidual
#	2: prStratified
#	3: prSystematic
#
# See: /docs/html/topic_resampling.html or http://mrpt.sourceforge.net/topic_resampling.html
#----------------------------------------------------------------------------------
resamplingMethod=3

sampleSize=15			// Sample size (for fixed number)
BETA=0.50			// Resampling ESS threshold


# ========================================================
#            MULTIMETRIC MAP CONFIGURATION
# See docs for (Google for) mrpt::maps::CMultiMetricMap
# ========================================================
# Creation of maps:
occupancyGrid_count=3
gasGrid_count=0
landmarksMap_count=0
beaconMap_count=0
pointsMap_count=0

# Selection of map for likelihood: (fuseAll=-1,occGrid=0, points=1,landmarks=2,gasGrid=3)
likelihoodMapSelection=-1

# Enables (1) / Disables (0) insertion into specific maps:
enableInsertion_pointsMap=1
enableInsertion_landmarksMap=1
enableInsertion_beaconMap=1
enableInsertion_gridMaps=1
enableInsertion_gasGridMaps=1

# ====================================================
#         CHybridMetricMapPDF::TPredictionParams
# ====================================================
powFactor=0.05					// A "power factor" for updating weights
pfAuxFilterOptimal_MaximumSearchSamples=200	// For PF algorithm=3

# -----------------------------------------------------------------
#   pfOptimalProposal_mapSelection
#  Only for PF algorithm=2 (Exact "pfOptimalProposal")
#   Select the map on which to calculate the optimal 
#    proposal distribution. Values:
#   0: Gridmap   -> Uses Scan matching-based approximation (based on Stachniss' work)
#   1: Landmarks -> Uses matching to approximate optimal
#   2: Beacons   -> Used for exact optimal proposal in RO-SLAM
# -----------------------------------------------------------------
pfOptimalProposal_mapSelection=2


# Adaptive sample size parameters ------------------
KLD_maxSampleSize=10000
KLD_minSampleSize=15
KLD_binSize_XY=10.00
KLD_binSize_PHI_deg=30
KLD_delta=0.01
KLD_epsilon=0.40


# ====================================================
#   MULTIMETRIC MAP: OccGrid #00
# ====================================================
# Creation Options for OccupancyGridMap 00:
[MappingApplication_occupancyGrid_00_creationOpts]
resolution=0.05
disableSaveAs3DObject=0


# Insertion Options for OccupancyGridMap 00:
[MappingApplication_occupancyGrid_00_insertOpts]
mapAltitude=0.30
useMapAltitude=1
maxDistanceInsertion=40
maxOccupancyUpdateCertainty=0.54
considerInvalidRangesAsFreeSpace=1
minLaserScanNoiseStd=0.001


# Likelihood Options for OccupancyGridMap 00:
[MappingApplication_occupancyGrid_00_likelihoodOpts]
likelihoodMethod=4		// 0=MI, 1=Beam Model, 2=RSLC, 3=Cells Difs, 4=LF_Trun, 5=LF_II
LF_decimation=5
LF_stdHit=0.10
LF_maxCorrsDistance=0.30
LF_zHit=0.999
LF_zRandom=0.001
LF_maxRange=30
LF_alternateAverageMethod=0


# Creation Options for OccupancyGridMap 01:
[MappingApplication_occupancyGrid_01_creationOpts]
resolution=0.05
disableSaveAs3DObject=0


# Insertion Options for OccupancyGridMap 01:
[MappingApplication_occupancyGrid_01_insertOpts]
mapAltitude=0.51
useMapAltitude=1
maxDistanceInsertion=40
maxOccupancyUpdateCertainty=0.54
considerInvalidRangesAsFreeSpace=1
minLaserScanNoiseStd=0.001


# Likelihood Options for OccupancyGridMap 01:
[MappingApplication_occupancyGrid_01_likelihoodOpts]
likelihoodMethod=4		// 0=MI, 1=Beam Model, 2=RSLC, 3=Cells Difs, 4=LF_Trun, 5=LF_II
LF_decimation=5
LF_stdHit=0.10
LF_maxCorrsDistance=0.30
LF_zHit=0.999
LF_zRandom=0.001
LF_maxRange=30
LF_alternateAverageMethod=0


# Creation Options for OccupancyGridMap 02:
[MappingApplication_occupancyGrid_02_creationOpts]
resolution=0.05
disableSaveAs3DObject=0


# Insertion Options for OccupancyGridMap 02:
[MappingApplication_occupancyGrid_02_insertOpts]
mapAltitude=1.47
useMapAltitude=1
maxDistanceInsertion=40
maxOccupancyUpdateCertainty=0.54
considerInvalidRangesAsFreeSpace=1
minLaserScanNoiseStd=0.001


# Likelihood Options for OccupancyGridMap 02:
[MappingApplication_occupancyGrid_02_likelihoodOpts]
likelihoodMethod=4		// 0=MI, 1=Beam Model, 2=RSLC, 3=Cells Difs, 4=LF_Trun, 5=LF_II
LF_decimation=5
LF_stdHit=0.10
LF_maxCorrsDistance=0.30
LF_zHit=0.999
LF_zRandom=0.001
LF_maxRange=30
LF_alternateAverageMethod=0

三个不同高度的水平激光扫描仪的三个网格地图

RBPF-SLAM算法(C ++库mrpt-slam)

使用RBPF进行网格映射的示例(rbpf-slam应用程序)

2.3。 pfOptimalProposal:网格地图的最佳建议
说明: 该 pfOptimalProposal 算法为栅格图对应于由G. Grisetti等人介绍了最佳提案的启发式近似。在:“用Rao-Blackwellized粒子滤波器进行网格映射的改进技术”,Grisetti,G. Stachniss,C. Burgard,W. IEEE TRANSACTIONS ON ROBOTICS(2007) - PDF MRPT中的实现依赖于版本的ICP格网格。另一种实现(独立于MRPT)是流行的 gmapping。 注意: pfOptimalProposal 需要设置参数 pfOptimalProposal_mapSelection (有关详细信息,请参阅示例配置文件)。 配置文件示例:

#------------------------------------------------------------
# Config file for the "RBPF-SLAM" application
# See: http://www.mrpt.org/list-of-mrpt-apps/application-rbpf-slam/
#------------------------------------------------------------


#=======================================================
# Section: [MappingApplication]
# Use: Here comes global parameters for the app.
#=======================================================
[MappingApplication]
# The source file (RAW-LOG) with action/observation pairs
rawlog_file=../../datasets/2006-01ENE-21-SENA_Telecom Faculty_one_loop_only.rawlog
rawlog_offset=0

# The directory where the log files will be saved (left in blank if no log is required)
logOutput_dir=LOG_MALAGA_RBPF_GRIDMAPPING_ICP_LevMar
LOG_FREQUENCY=10			// The frequency of log files generation:
SAVE_MAP_IMAGES=1
SAVE_3D_SCENE=1
SAVE_POSE_LOG=0

CAMERA_3DSCENE_FOLLOWS_ROBOT=0
SHOW_PROGRESS_IN_WINDOW=1

insertionLinDistance     = 1.20	  // The distance threshold for inserting observations in the map (meters)
insertionAngDistance_deg = 35.0	  // The distance threshold for inserting observations in the map (degrees)

localizeLinDistance		 = 1.20	  // The distance threshold for updating the RBPF
localizeAngDistance_deg  = 35.0	  // The distance threshold for inserting observations in the map (degrees)

#----------------------------------------------------------------------------------
# The Particle Filter algorithm:
#	0: pfStandardProposal
#	1: pfAuxiliaryPFStandard
#	2: pfOptimalProposal      *** (ICP-based (Grisetti method),...)
#	3: pfAuxiliaryPFOptimal	  *** (Optimal SAMPLING)
#
# See: http://babel.isa.uma.es/mrpt/index.php/Particle_Filter_Algorithms
#----------------------------------------------------------------------------------
PF_algorithm		= 2
adaptiveSampleSize	= 0		// 0: Fixed # of particles, 1: KLD adaptive

#----------------------------------------------------------------------------------
# The Particle Filter Resampling method:
#	0: prMultinomial
#	1: prResidual
#	2: prStratified
#	3: prSystematic
#
# See: http://babel.isa.uma.es/mrpt/index.php/Resampling_Schemes
#----------------------------------------------------------------------------------
resamplingMethod=0

sampleSize=10			// Sample size (for fixed number)
BETA=0.50			// Resampling ESS threshold


# ========================================================
#            MULTIMETRIC MAP CONFIGURATION
# See docs for (Google for) mrpt::maps::CMultiMetricMap
# ========================================================
# Creation of maps:
occupancyGrid_count  = 1
gasGrid_count        = 0
landmarksMap_count   = 0
beaconMap_count      = 0
pointsMap_count      = 0

# Selection of map for likelihood: (fuseAll=-1,occGrid=0, points=1,landmarks=2,gasGrid=3)
likelihoodMapSelection=-1


# ====================================================
#         CHybridMetricMapPDF::TPredictionParams
# ====================================================
#powFactor=0.01					// A "power factor" for updating weights
pfAuxFilterOptimal_MaximumSearchSamples=250	// For PF algorithm=3


# -----------------------------------------------------------------
#   pfOptimalProposal_mapSelection
#  Only for PF algorithm=2 (Exact "pfOptimalProposal")
#   Select the map on which to calculate the optimal 
#    proposal distribution. Values:
#   0: Gridmap   -> Uses Scan matching-based approximation (based on Stachniss work)
#   1: Landmarks -> Uses matching to approximate optimal
#   2: Beacons   -> Used for exact optimal proposal in RO-SLAM
#   3: Points-map-> Uses Scan matching-based approximation (based on Stachniss work)
# -----------------------------------------------------------------
pfOptimalProposal_mapSelection=0

# Adaptive sample size parameters ------------------
KLD_maxSampleSize=150
KLD_minSampleSize=15
KLD_binSize_XY=0.03
KLD_binSize_PHI_deg=1
KLD_delta=0.001
KLD_epsilon=0.5


# ====================================================
#         ICP configuration.
#      Only used if PF_algorithm=2
# ====================================================
# 0: icpClassic
# 1: icpLevenbergMarquardt
# 2: icpIKF
ICP_algorithm    = 0
maxIterations    = 60 		// The maximum number of iterations to execute if convergence is not achieved before
thresholdDist    = 1.00		// Initial maximum distance for matching a pair of points
thresholdAng_DEG = 5		// An angular factor (in degrees) to increase the matching distance for distant points.
ALFA             = 0.50 			// After convergence, the thresholds are multiplied by this constant and ICP keep running (provides finer matching)
smallestThresholdDist      = 0.10// This is the smallest the distance threshold can become after stopping ICP and accepting the result.
onlyClosestCorrespondences = 1	// 1: Use the closest points only, 0: Use all the correspondences within the threshold 

# ====================================================
#   MULTIMETRIC MAP: OccGrid #00
# ====================================================
# Creation Options for OccupancyGridMap 00:
[MappingApplication_occupancyGrid_00_creationOpts]
resolution=0.07
disableSaveAs3DObject=0


# Insertion Options for OccupancyGridMap 00:
[MappingApplication_occupancyGrid_00_insertOpts]
mapAltitude					= 0
useMapAltitude				= 0
maxDistanceInsertion		= 35
maxOccupancyUpdateCertainty	= 0.60
considerInvalidRangesAsFreeSpace = 1
minLaserScanNoiseStd		= 0.001
wideningBeamsWithDistance 	= 0


# Likelihood Options for OccupancyGridMap 00:
[MappingApplication_occupancyGrid_00_likelihoodOpts]
likelihoodMethod=4		// 0=MI, 1=Beam Model, 2=RSLC, 3=Cells Difs, 4=LF_Trun, 5=LF_II
LF_decimation=5
LF_stdHit=0.50
LF_maxCorrsDistance=0.30
LF_zHit=0.999
LF_zRandom=0.001
LF_maxRange=30
LF_alternateAverageMethod=0

enableLikelihoodCache=1

使用中等大小的闭环网格进行网格映射。
2.4。 pfOptimalProposal:点图的最佳建议
说明: 该 pfOptimalProposal 算法点地图是Grisetti等人的想法的一个版本,而是适用于点云(更快的更新比栅格图)。AFAIK,这种方法尚未在文献中发表,但工作很好。 注意: pfOptimalProposal 需要设置参数 pfOptimalProposal_mapSelection (有关详细信息,请参阅示例配置文件)。 配置文件示例:

#------------------------------------------------------------
# Config file for the "RBPF-SLAM" application
# See: http://www.mrpt.org/list-of-mrpt-apps/application-rbpf-slam/
#------------------------------------------------------------
# Optimal sampling algorithm: 3: pfAuxiliaryPFOptimal


#=======================================================
# Section: [MappingApplication]
# Use: Here comes global parameters for the app.
#=======================================================
[MappingApplication]
# The source file (RAW-LOG) with action/observation pairs
rawlog_file=../../datasets/2006-01ENE-21-SENA_Telecom Faculty_one_loop_only.rawlog
rawlog_offset=0

# The directory where the log files will be saved (left in blank if no log is required)
logOutput_dir=LOG_MALAGA_RBPF_GRIDMAPPING_ICP_LevMar
LOG_FREQUENCY=10			// The frequency of log files generation:
SAVE_MAP_IMAGES=1
SAVE_3D_SCENE=1
SAVE_POSE_LOG=0

CAMERA_3DSCENE_FOLLOWS_ROBOT=0
SHOW_PROGRESS_IN_WINDOW=1

insertionLinDistance     = 1.20	  // The distance threshold for inserting observations in the map (meters)
insertionAngDistance_deg = 35.0	  // The distance threshold for inserting observations in the map (degrees)

localizeLinDistance		 = 1.20	  // The distance threshold for updating the RBPF
localizeAngDistance_deg  = 35.0	  // The distance threshold for inserting observations in the map (degrees)

# Console verbosity level: 
# One of: DEBUG, INFO (default), WARN, ERROR
verbosity_level = DEBUG

#----------------------------------------------------------------------------------
# The Particle Filter algorithm:
#	0: pfStandardProposal
#	1: pfAuxiliaryPFStandard
#	2: pfOptimalProposal      *** (ICP-based (Grisetti method),...)
#	3: pfAuxiliaryPFOptimal	  *** (Optimal SAMPLING)
#
# See: http://babel.isa.uma.es/mrpt/index.php/Particle_Filter_Algorithms
#----------------------------------------------------------------------------------
PF_algorithm		= 2
adaptiveSampleSize	= 0		// 0: Fixed # of particles, 1: KLD adaptive

#----------------------------------------------------------------------------------
# The Particle Filter Resampling method:
#	0: prMultinomial
#	1: prResidual
#	2: prStratified
#	3: prSystematic
#
# See: http://babel.isa.uma.es/mrpt/index.php/Resampling_Schemes
#----------------------------------------------------------------------------------
resamplingMethod=0

sampleSize=10			// Sample size (for fixed number)
BETA=0.50			// Resampling ESS threshold


# ========================================================
#            MULTIMETRIC MAP CONFIGURATION
# See docs for (Google for) mrpt::maps::CMultiMetricMap
# ========================================================
# Creation of maps:
occupancyGrid_count  = 0
gasGrid_count        = 0
landmarksMap_count   = 0
beaconMap_count      = 0
pointsMap_count      = 1

# Selection of map for likelihood: (fuseAll=-1,occGrid=0, points=1,landmarks=2,gasGrid=3)
likelihoodMapSelection=-1


# ====================================================
#         CHybridMetricMapPDF::TPredictionParams
# ====================================================
#powFactor=0.01					// A "power factor" for updating weights
pfAuxFilterOptimal_MaximumSearchSamples=250	// For PF algorithm=3


# -----------------------------------------------------------------
#   pfOptimalProposal_mapSelection
#  Only for PF algorithm=2 (Exact "pfOptimalProposal")
#   Select the map on which to calculate the optimal 
#    proposal distribution. Values:
#   0: Gridmap   -> Uses Scan matching-based approximation (based on Stachniss work)
#   1: Landmarks -> Uses matching to approximate optimal
#   2: Beacons   -> Used for exact optimal proposal in RO-SLAM
#   3: Points-map-> Uses Scan matching-based approximation (based on Stachniss work)
# -----------------------------------------------------------------
pfOptimalProposal_mapSelection=3

# Adaptive sample size parameters ------------------
KLD_maxSampleSize=150
KLD_minSampleSize=15
KLD_binSize_XY=0.03
KLD_binSize_PHI_deg=1
KLD_delta=0.001
KLD_epsilon=0.5


# ====================================================
#         ICP configuration.
#      Only used if PF_algorithm=2
# ====================================================
# 0: icpClassic
# 1: icpLevenbergMarquardt
# 2: icpIKF
ICP_algorithm    = 0
maxIterations    = 60 		// The maximum number of iterations to execute if convergence is not achieved before
thresholdDist    = 1.00		// Initial maximum distance for matching a pair of points
thresholdAng_DEG = 5		// An angular factor (in degrees) to increase the matching distance for distant points.
ALFA             = 0.50 			// After convergence, the thresholds are multiplied by this constant and ICP keep running (provides finer matching)
smallestThresholdDist      = 0.10// This is the smallest the distance threshold can become after stopping ICP and accepting the result.
onlyClosestCorrespondences = 1	// 1: Use the closest points only, 0: Use all the correspondences within the threshold 

# ====================================================
#   MULTIMETRIC MAP: PointsMap #00
# ====================================================
# Likelihood Options for Pointsmap 00:
[MappingApplication_pointsMap_00_likelihoodOpts]
max_corr_distance	= 1.0
sigma_dist			= 0.05

用中等尺寸环路闭合点云映射。

weinxin
微信公众号
agvba是一个分享AGV知识和agv案例视频的网站。