mapper not generating expected occupancy grid map

[Nitin]

Hi,

I have been running a few experiments with 2D lidar data. I found that mapper do not generates complete occupancy grid initially. I am attaching two pictures. Please see that a lot of area is left as unknown in the map although, laser beams has passed through them.

This picture shows starting position of the robot. Size of a cell in rviz display is 0.25x0.25 m-sq while size of a cell in Stage grid is 1m-sq. Laser range in simulation is 10meters.

Further, Robot starts moving backward. Next picture is output after some movement backward.

I have been using following configuration data

Mapper

robot_id: 1
grid_resolution: 0.25
range_threshold: 10.0 #Maximum range of laser sensor
map_update_rate: 1
publish_pose_graph: true
max_covariance: 0.01 # Threshold to accept result of particle filter
transform_publish_period: 0.1
min_map_size: 20 #minimum number of nodes in map for localization
#laser_topic: base_scan
#map_frame: /map
#map_topic: map
#laser_frame: base_laser_link
#robot_frame: base_link
#odometry_frame: odom1
#offset_frame: offset

Localizer

min_particles: 2500
max_particles: 10000

Karto

MinimumTravelDistance: 1.0
MinimumTravelHeading: .52
LoopSearchMaximumDistance: 5.0
ScanBufferMaximumScanDistance: 20.0
ScanBufferSize: 8

For a full list of available parameters and their

corresponding default values, see OpenKarto/OpenMapper.h

#UseScanMatching: true
#UseScanBarycenter: true
#MinimumTravelDistance: 1.0
#MinimumTravelHeading: 0.52
#ScanBufferSize: 70
#ScanBufferMaximumScanDistance: 20
#UseResponseExpansion: false

Variance of penalty for deviating from odometry when scan-matching. Use lower values if the robot's odometer is more accurate.

#DistanceVariancePenalty: 0.6
#MinimumDistancePenalty: 0.5
#AngleVariancePenalty: 0.1225 # variance of 20 x 20 degree sq in radians
#MinimumAnglePenalty: 0.9
#LinkMatchMinimumResponseFine: 0.6
#LinkScanMaximumDistance: 5.0
#CorrelationSearchSpaceDimension: 0.3 #Seach for correlation in area of 30x30 cm-square
#CorrelationSearchSpaceResolution: 0.1
#CorrelationSearchSpaceSmearDeviation: 0.03
#CoarseSearchAngleOffset: 0.34906585 #20 degrees
#FineSearchAngleOffset: 0.00349 #0.2degrees
#CoarseAngleResolution: 0.0349 #2 degrees
#LoopSearchSpaceDimension: 8 #search area of size 8x8 meter square
#LoopSearchSpaceResolution: 0.05
#LoopSearchSpaceSmearDeviation: 0.03
#LoopSearchMaximumDistance: 10.0
#When the loop closure detection finds a candidate it must be part of a large set of linked scans. If the chain of scans is less than this value, we do not attempt to close the loop.
#LoopMatchMinimumChainSize: 10
#LoopMatchMaximumVarianceCoarse: 0.16 #The co-variance values for a possible loop closure have to be less than this value to consider a viable solution
#LoopMatchMinimumResponseCoarse: 0.7 #if probability value is larger than this, initiate coarser level search
#LoopMatchMinimumResponseFine: 0.7 #if probability value is larger than this, initiate finer level search

###########################################################

Any help in how can i refine output of mapping will be appreciated.

[Sebastian Kasperski]

Is it intentional to use such a low resolution for the map? I would opt for a much higher one for indoor mapping (e.g. 0.05). And try to increase the mapping range_threshold as well. Maybe the rest of your points is out of this range.

[Nitin]

Yes, low resolution use was intentional to enable faster testing and debug. I agree 5cm resolution is better choice for experiments on robots. I realized that output is influenced by the max range of sensor. If I set ranger's max range value to higher values (10 or 20 meter), the output from nav2d_karto is much better. otherwise, there is tendency to leave out those rays which do not hit any obstacles i.e. range reading is max range value. This behaviour I have also observed in costmap2d_ros. Costmap2d_ros do not consider those laser beams for which reading is maximum value. Rest of the points are not out of range of laser sensor beams although, still grid cells are not clear.

…

On Thu, Jun 1, 2017 at 5:00 PM, Sebastian Kasperski < ***@***.***> wrote: Is it intentional to use such a low resolution for the map? I would opt for a much higher one for indoor mapping (e.g. 0.05). And try to increase the mapping range_threshold as well. Maybe the rest of your points is out of this range. — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub <#33 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AHp19_axhshTENZQTAOjHaRD1oR1gbPVks5r_qC8gaJpZM4NssDF> .

[Sebastian Kasperski]

I am not sure I fully understand the problem. The pictures look okay to me, given the shown laser scans and the low map resolution. Which cells should be cleared, that are not?

[Nitin]

I realised that problem can be avoided using a finer resolution or long range sensor, at least in simulations.