NOTE: This package is unreleased at this point, the documentation is for Willow Garage internal use only.

NOTE: At this point, this package has been superseded by collider.

Overview

This package contains the collision_octomap_node node, which builds a 3D representation of sensed data, much like collision_map. In fact, this node is designed to be a "drop-in" replacement for collision_map_node. The difference between the two is that collision_octomap_node maintains its internal representation as an octree of free and occupied space, using the octomap2 library. This allows the generation collision maps in a large volume, whereas the implementation of collision_map only maintains a collision map in the space currently visible to the robot's sensors.

Sensor Sources

The sensor sources can be any point cloud of type PointCloud. However, for reasons explained below, sources that contain a very large number of points and update relatively rapidly, such as stereo vision points, should not be used. The overall point cloud insertion time for collision_octomap_node is much longer than for collision_map because it uses the octree representation, which needs to clear out freespace as well as inserting occupied cells. If large, rapidly updating point clouds need to be temporarily incorporated into the collision map (say, for monitoring of an arm with the stereo vision), one could add a small modification to the code to allow overlaying collision boxes from a given point source without actually inserting them into the octree.

Development Note

collision_octomap_node publishes the exact same type of the 3D collision map as collision_map, so its output can be used directly by other nodes for collision checking. However, collision_octomap_node will typically publish its collision map in a world frame (odom_combined, map, etc.) whereas collision_map publishes in the base_link frame. While testing collision_octomap_node, I noticed that some of the other nodes that process the collision map would occasionally time out trying to convert the collision map from the world frame, causing poor performance for things like arm collision checking. I'm not sure if this was because I was heavily loading the system with other processes as well, but it's something to look out for.

ROS API

ROS Parameters

Parameters

~fixed_frame (string, default: base_link)
  • The frame used to maintain the internal octree representation of the collision map, usually some sort of world frame
~resolution (double, default: 0.1)
  • The smallest resolution of the internal octree, in meters. Has a large effect on insertion speed. Not recommended to use less than 0.02, at least with the current implementation of octomap2.
~max_range (double, default: -1.0)
  • The maximum beam length that will be inserted into the octree. If a point is further than this distance, its only effect will be to clear out freespace. The default of -1.0 means the full beam length.
~pruning_period (int, default: 5)
  • The internal octree will be pruned after every x-th data insertion, where x is the pruning period. There is a tradeoff between pruning too often and not often enough that needs to be found.
~dimension_x (dimension_y, dimension_z) (double, default: 0.0)
  • The x (y, z) dimension of a workspace. Sensor points outside this workspace will not be inserted into the octree, which can cause problems when it comes to clearing out points from inside the workspace. Not recommended to be used in its current implementation. Default value of 0.0 means workspace is infinitely large (recommended).
~origin_x (origin_y, origin_z) (double, default: 0.0)
  • The x (y, z) origin of a workspace in the fixed frame. Sensor points outside this workspace will not be inserted into the octree, which can cause problems when it comes to clearing out points from inside the workspace. Not recommended to be used in its current implementation.
cloud_sources (XmlRpcValue, default: none)
  • The cloud point sources which will be used as inputs to collision_octomap_node, specified in a YAML file. Details below.

cloud_sources YAML file

An example of the cloud_sources YAML file is given below. Note that this is the same format used for the regular collision_map.

cloud_sources:
  - name: full_cloud_filtered
    frame_subsample: 1
    point_subsample: 1
    sensor_frame: laser_tilt_mount_link

Published Topics

Data Topics
collision_map_out (mapping_msgs/CollisionMap)
  • The output collision map, defined as a set of boxes with a given center point and edge lengths.
point_cloud_out (sensor_msgs/PointCloud)
  • The output collision map in point cloud form, which is simply the center points of the collision map boxes. Used for nodes that need a point cloud input topic, but should probably be eliminated in the future.
Visualization Topics
octomap_insertion_info_array (visualization_msgs/MarkerArray)
  • Text information about the latest insertion.
octomap_occupied_marker_array (visualization_msgs/MarkerArray)
  • Box markers of the current occupied cells in the octree. Useful for debugging.
octomap_freespace_marker_array (visualization_msgs/MarkerArray)
  • Box markers of the current freespace cells in the octree. Useful for debugging.

Example Usage

The collision_octomap_node should be launched the same way as as collision_map_node. An example launch file is shown below.

<launch>

  <node pkg="collision_octomap" type="collision_octomap_node" name="collision_octomap_node" respawn="true" output="screen" machine="c2">

    <param name="fixed_frame" type="string" value="map" />
    <param name="resolution" type="double" value="0.025" />
    <param name="max_range" type="double" value="2.5" />
    <param name="pruning_period" type="int" value="10" />

    <remap from="collision_map_out" to="collision_octomap_occ" />
    <remap from="point_cloud_out" to="collision_point_cloud" />

    <!-- cloud sources -->
    <rosparam command="load" file="$(find pr2_arm_navigation_perception)/config/collision_map_sources.yaml" />  
  </node>
</launch>

In the above launch file, we are using the "map" fixed frame, which is provided by the navigation stack. Make sure you launch some flavor of the navigation stack if you wish to use the map frame, otherwise use something like "odom_combined". If you want to insert laser snapshots (recommended) instead of individual laser scans, you should configure your assembler with the same fixed frame used by collision_octomap_node. Example launch file below.

<launch>
  <!-- assemble pointcloud into a full world view -->
  <node pkg="laser_assembler" type="point_cloud_assembler" output="screen"  name="point_cloud_assembler">
    <remap from="cloud" to="tilt_scan_cloud_filtered"/>
    <param name="tf_cache_time_secs" type="double" value="10.0" />
    <param name="tf_tolerance_secs" type="double" value="0.0" />
    <param name="max_clouds" type="int" value="400" />
    <param name="ignore_laser_skew" type="bool" value="true" />
    <param name="fixed_frame" type="string" value="map" />
  </node>
  
  <node pkg="pr2_arm_navigation_perception" type="pr2_laser_snapshotter" output="screen" name="laser_snapshotter">
    <remap from="laser_scanner_signal" to="laser_tilt_controller/laser_scanner_signal"/>
    <remap from="build_cloud" to="point_cloud_assembler/build_cloud" />
    <remap from="full_cloud" to="full_cloud_filtered" />
  </node> 
</launch>

Wiki: collision_octomap (last edited 2012-03-19 10:42:43 by FelixKolbe)