About | Support | Discussion Forum | Index | Service Status | ros @ Robotics Stack Exchange

Advanced Parameter Tuning

Description: This tutorial tells you which parameter to change to improve performances

Keywords: parameters

Tutorial Level: INTERMEDIATE

Introduction

This tutorial is aimed at helping people using rtabmap in a more advanced way. Only after you have successfully installed rtabmap and rtabmap_ros, you shall start this tutorial.

Change Parameters

To know what parameters you can change, and what do they mean, you can use rtabmap gui and check the Preferences dialog or you can do:

rosrun rtabmap_ros rtabmap --params
rosrun rtabmap_ros rgbd_odometry --params

If you want to know a specific parameter, for example about g2o

$ rosrun rtabmap_ros rtabmap --params | grep g2o
Param: Optimizer/Robust = "false" [Robust graph optimization using Vertigo (only work for g2o and GTSAM optimization strategies). Not compatible with "RGBD/OptimizeMaxError" if enabled.]
Param: Optimizer/Strategy = "2"   [Graph optimization strategy: 0=TORO, 1=g2o and 2=GTSAM.]
Param: g2o/Optimizer = "0"        [0=Levenberg 1=GaussNewton]
Param: g2o/PixelVariance = "1"    [Pixel variance used for SBA.]
Param: g2o/Solver = "0"           [0=csparse 1=pcg 2=cholmod]

There are generally three way of changing parameters.

1) Use a config.ini file.

The default config.ini should be automaticlly generated once you launch rtabmap through terminal by "$ rtabmap" command. Its location is ~/Documents/RTAB-Map/config.ini. You can copy this config.ini file into your own package and change your launch file cfg parameter into the corresponding location.

For example, you can set Odom\Strategy=1 in config.ini, then in a package called arti_vision, you can put the config.ini in the config folder and renamed it into rtabmap.ini. Calling rtabmap.launch with the cfg argument:

<arg name="rtabmap_args" default="$(find arti_vision)/config/rtabmap.ini" />

2) Use <param> tag

This is the normal ROS way of reading parameters. For exmaple, you can put this under rgbd_odometry node

<param name="Odom/Strategy" value="0"/>

3) Add parameters in arguments of the node.

If you just want to temporarily play with some parameters, you can quickly put it under the rtabmap_args argument:

<arg name="rtabmap_args" default="--Odom/Strategy 0"/> 

All three way of changing parameters have the same effects. Choose whatever you want, though the second approach is the standard way to set parameters in ROS.

Visual Odometry

Tuning visual odometry parameter is very important, it directly affects how good RTAB-Map generally performs.

Increase Speed

If you have a very limited computing power, you probably want to play with the following parameters, so that you can increase the odometry frequency, and thus, the robot won't get lost. Here are some parameters you should try:

Change the Visual Odometry Stategy into Frame to Frame

<!-- 0=Frame-to-Map (F2M) 1=Frame-to-Frame (F2F) -->
<param name="Odom/Strategy" value="1"/>

Change the Matching Correspondences into Optical Flow

Optical flow may give more matches, but less robust correspondences:

<!-- Correspondences: 0=Features Matching, 1=Optical Flow -->
<param name="Vis/CorType" value="1"/>

Reduce the Maximum Features

<!-- maximum features map size, default 2000 -->
<param name="OdomF2M/MaxSize" type="string" value="1000"/> 
<!-- maximum features extracted by image, default 1000 -->
<param name="Vis/MaxFeatures" type="string" value="600"/>

Increase the Frame Rate of Camera and Reduce the Resolution

Those two aspects are very effective, increase your camera's frame rate directly affects how fast the VO can track your robot. Reduce resolution can greatly increase the speed of processing but may reduce the accuracy. You need to make a good balance based on your hardware and use case.

Change Feature Distance

If your images are 720p or more, you may want to increase GFTT distance (minimum distance between extracted features) so that you have more distributed features in the images:

<node pkg="rtabmap_ros" type="rgbd_odometry" name="rgbd_odometry">
        <param name="GFTT/MinDistance" type="string" value="10"/> <!-- default 5 pixels -->
</node>

Odometry Auto-Reset

When calling reset_odom service, rtabmap will start a new map, hiding the old one until a loop closure is found with the previous map. If you don't want rtabmap to start a new map when odometry is reset and wait until a first loop closure is found, you can set Rtabmap/StartNewMapOnLoopClosure to true. If parameter Odom/ResetCountdown is set to 1 (default 0=disabled), odometry will automatically reset one frame after being lost, i.e., it has the same effect than calling reset_odom service.

<node pkg="rtabmap_ros" type="rgbd_odometry" name="rgbd_odometry">
        <param name="Odom/ResetCountdown"              value="1" />
</node>
<node pkg="rtabmap_ros" type="rtabmap" name="rtabmap">
        <param name="Rtabmap/StartNewMapOnLoopClosure" value="true"/>
</node>

Mapping

Three DOF mapping

For most of UGV, the vehicle only runs on a flat ground, in this way, you can force the visual odometry to track the vehicle in only 3DOF (x,y,theta) and increase the robustness of the map. For rtabmap, we can also constraint to 3 DoF loop closure detection and graph optimization:

<node pkg="rtabmap_ros" type="rgbd_odometry" name="rgbd_odometry">
        <param name="Reg/Force3DoF"    value="true" />
</node>

<node pkg="rtabmap_ros" type="rtabmap" name="rtabmap">
        <param name="Reg/Force3DoF"    value="true" />
        <param name="Optimizer/Slam2D" value="true" />
</node>

Change inital costmap size for move_base

You can initialize the minimum size of the map with grid_size parameter:

<node pkg="rtabmap_ros" type="rtabmap" name="rtabmap">
   <param name="grid_size" type="double" value="50"/> <!-- 50 meters wide -->
</node>

Reduce Point Cloud Noises

<param name="cloud_noise_filtering_radius" value="0.05"/>
<param name="cloud_noise_filtering_min_neighbors" value="2"/>

Reference: pcl::RadiusOutlierRemoval()

Change Projected Occupancy Grid Characteristic

proj_max_ground_angle means mapping maximum angle between point's normal to ground's normal to label it as ground. Points with higher angle difference are considered as obstacles.

Increasing proj_max_ground_angle will make the algorithm include points with normal's angle farther from z+ axis as ground.

<param name="proj_max_ground_angle" value="45"/>

proj_max_ground_height means mapping maximum height of points used for projection. Used for proj_map published topic.

Increasing the proj_max_ground_height will make the algorithm ignore points that are under this threshold while projection. All points below this threshold will be ignored:

<param name="proj_max_ground_height" value="0.1"/>

proj_max_height means mapping maximum height of points used for projection:

<param name="proj_max_height" value="2.0"/>

proj_min_cluster_size means mapping minimum cluster size to project the points. The distance between clusters is defined by 2*grid_cell_size:

<param name="proj_min_cluster_size" value="20"/>