Cognitive Perception Server Perception interface to highlevel planning (kipla)
Cognitive Perception (CoP)
Organizes and connects perception mechanisms.
+++Installation: How to install CoP:+++
Get the current lastest version of cognitive
svn co https://tum-ros-pkg.svn.sourceforge.net/svnroot/tum-ros-pkg tum-ros-pkg
Add the folder you check out tum-ros-pkg to your ROS_PACKAGE_PATH.
++ If you have HALCON ++
If you have HALCON (http://www.halcon.com) you can use all plugins defined in cop_halcon_plugin (including Camera drivers, CAD matching, Template matching, color classification, ...), otherwise only cop_sr4_plugin and simple_plugins are available.
Make sure the your $HALCONROOT ist set correctly to the installation directory of halcon and the $HALCONARCH spcifies your system (usually "x64-linux2.4-gcc40").
++ If you DO NOT have HALCON ++
Do the following:
# roscd cop_halcon_plugin
# touch ROS_BUILD_BLACKLIST
Always:
Do the following:
# rosmake cop
cop is a package that loads an example of a configuration for cop from http://www9.in.tum.de/~klank/cop_resource.tar.gz and this package depends on cognitive perception, locate object server and all plugins.
Now everything should be able to run.
Additionally you can build the cop_rviz_plugin with
# rosmake cop_rviz_plugin
++ Testing (this example works only with HALCON) ++
Download
http://www9.in.tum.de/~klank/trial1_2010-02-02-18-03-51.bag
Then open 4 consoles and start the 4 programs (don't use launch files to get a better feeling what is happening). Start the following programs:
1# roscore
2# rosrun jlo jlo
3# rosplay trial1_2010-02-02-18-03-51.bag
4# roscd cop/resource; ../../cognitive_perception/cop_srv remote_cop_config.ini
Additionally you could start rviz and add a new display CopAnswerDiplay and subscribe to the default topic of test_client::test_cop "/tracking/out"
Now you could see, if you additionally subscribe to /tf and a camera (e.g. /cop/rightcamera) how our robot drives through our kitchen and looks at a table.
Remeber to stop the rosplay at a certain point when you see the table.
+++Usage: How to call cop +++
To show the basic functionality of CoP we will use the debug client for the cop_srv, which is in the package test_client:
# rosmake test_client
NOTICE: CoP is not based on tf but on the located object service (jlo), which is aware of all tf transformation, because it works with Position covariances. So, we first call jlo to find out which id was assigned to the swissranger:
# rosrun test_client query_jlo /sr4 This will result in something like:
# Showing PosId 147 (/sr4) with parent 143:
#-0.999952 0.000413 0.009839 -0.061978
#-0.000207 -0.999782 0.020872 0.001763
#0.009845 0.020869 0.999734 -0.037964
#0.000000 0.000000 0.000000 1.000000
#Cov:
#0.000000 0.000000 0.000000 0.000000 0.000000 0.000000
#0.000000 0.000000 0.000000 0.000000 0.000000 0.000000
#0.000000 0.000000 0.000000 0.000000 0.000000 0.000000
#0.000000 0.000000 0.000000 0.000000 0.000000 0.000000
#0.000000 0.000000 0.000000 0.000000 0.000000 0.000000
#0.000000 0.000000 0.000000 0.000000 0.000000 0.000000
#Type: 1
#pos dist: 0.072703
In this case the swissranger was assigned to 147.
++ Understand test_cop: ++
# rosrun test_client test_cop
This should result in somwthing like this:
# ClassName can be Cluster, Mug, IceTea, black, white, AssamBlend, ...
# LoID is a search space (use querylo to resolve tf-names to LoIDs)
# NumObjects specifies the maximum number of objects returned
# Command Defaul is 0:=Locate 256:=Track 768:=Refine 2048:=StopTrack
++ CoP Secondary Search space generation: ++
cop can use PlaneClusters (thanks at Radu Rusu and Dejan Pangercic) to create search spaces, use this functionality to also test the visualization in rviz:
# rosrun test_client test_cop Cluster 147 5
This should result in something like this:
#Called cop
#got answer from cop! (Errors: )
#Found Obj nr 1050 with prob 1.000000 at pos 328
# Cluster
#Found Obj nr 1053 with prob 0.375639 at pos 330
# Cluster
#Found Obj nr 1056 with prob 0.097121 at pos 329
# Cluster
#Found Obj nr 1059 with prob 0.082478 at pos 331
# Cluster
#Found Obj nr 1062 with prob 0.067570 at pos 332
# Cluster
#End!
This created new object instances in cop with object indices (here the column with 1050, 1053, 1056, 1059, 1062) Those objects are located at the LOIDs (like 147 of our swissranger), which are in the last column (with the values here 328, 330, 329, 331 and 332).
++ CoP Object Refinement: ++
Cop can now learn properties of those clusters. Tell cop to "Refine" one of the clusters by doing this:
# rosrun test_client test_cop 1050 328 1 768
With the object id and the corresponding position you got by the last call. 1 means its one object and 768 is the command for Refine (=0x300).
This should result in something like:
# Called cop
# got answer from cop! (Errors: )
# Found Obj nr 1050 with prob 0.000000 at pos 328
# Cluster black Texture_1066
# End!
which means that you learned now have learned two additional properties: black and Texture_1066. Those are references to a color classification result and a newly learned texture.
++ CoP Locate a specific object: ++
# rosrun test_client test_cop 1050
+++ API: Interfaces that can be used by plugins +++
The following plugins can be implemented and can be loaded by CoP.