Configuring cop

The package cognitive_perception build a library called cop and a service called cop_srv.

The package cop contains examples of configurations for this service.

The configuration is a XML-file that is passed as the first command line parameter to the cop_srv and starts a set of rosservices in the node named by the second command line parameter (careful with ros launch files here).

A cop config is a xml file. it contains a <config> node with the following sub elements:
. <VisualFinder/>: Configures algorithms used for localizing special objects or concepts.
. <VisLearner/>: Configures algorithms used to learn new Descriptors or to classify existing objects
. <ImageInputSystem/>: Configures the sensor setup loaded on start up
. <SignatureDB/> Configures the available Descriptors on start up
. <AttentionManager/>:

An example might look like:

<?xml version="1.0" encoding="ISO-8859-1"?>
  <config>
    <VisualFinder>
      ...
    </VisualFinder>
    <VisLearner>
      ...
    </VisLearner>
    <ImageInputSystem>
     ...
    </ImageInputSystem>
    <SignatureDB>
     ...
    </SignatureDB>
    <AttentionManager />
  </config>

Using the /cop/in Service

The main functionality of cop can be used by calling the service at /[cop]/in It is of type vision_srvs/cop_call which looks like this (rossrv show cop_call):

string outputtopic
string[] object_classes
uint64[] object_ids
uint64 action_type
uint64 number_of_objects
vision_msgs/apriori_position[] list_of_poses
  float64 probability
  uint64 positionId
---
uint64 perception_primitive

The parameter outputtopic specifies the topic on which the answers are expected. This topic will be advertised by cop, and cop will wait a while for the caller of this service to subscribe to it, before publishing the first result.

The parameter object_class can take a list of semantic classes that selects the models that may be used for the specified action. If one of the given names is not yet registered as a class, it is considered as a new annotation for all given ids.

The parameter object_ids can take a list of objects that selects the models and signatures that can be used for the requested action.

The parameter action_type specifies whether this action is thought to Localize an object, Track it, Refine (= Learning new properties and classifying it) it, or just Look it up.

The parameter number_of_objects specifies the maximum numbers of objects that is returned by a the action.

The parameter list_of_poses contains at least one position where the object has to expected. If this value is empty, all actions will fail, but the Locate action that will try to create candidate regions by a geometric segmentation, if available.

Using the /cop/save Service

Next to the default service cop_call cop provides an additional service: /cop/save

This service is thought to provide a possibility to share some learned models with other robots. It allows to send an ObjectID (see <SignatureDB/>) and the result will be a local path of XML-file saving the necessary parameters. Additionally there is a list of relative paths returned which contains all files referred in the XML-file that are necessary to create the model.

There exist a test client: Given a running instance of cop: rosrun test_client save_cop 2

will return something like: Called cop XMLFilename: /work/klank/src/tumros-internal/perception/cop/resource/signature1280824708.xml Additional filename 0: plate_brownrings.DXF