Running a pipeline in RoboSherlock¶
The demo provided is very simple and is meant to introduce users to the components of the framework, and
how to use them. Download the provided
sample bag file. The bagfile was recorded using
a PR2 robot and contains a short stream of kinect data, namely the topics:
/kinect_head/rgb/image_color/compressed /kinect_head/depth_registered/comressedDepth /kinect_head/rgb/camera_info /tf
Tf is needed to get the transformation between camera frame and map or robot base. This feature can be turned off in the camera configuration files.
Launch the robosherlock demo:
roslaunch robosherlock rs.launch
The launch file will initialize active components of RoboSherlock and will wait for data to be published on the camera topics. The launch files sets some default params (like the analysis engine to run by default). These can be changed through rosparams, e.g.:
roslaunch robosherlock rs.launch ae:=<my_ae_name>
Start the bagfile:
rosbag play test.bag --loop
You should see the results of the individual annotators in the visualizer windows (one for the colored image and one for the generated point clouds). Pressing left or right will change the view. You should see an output similar to the one below:
The demo is a very simple example of how perception pipelines are defined and run in RoboSherlock. Launching the pipeline happens through running the run executable, which takes as parameter the name of the aggregate analysis engine, in this case demo. The definition of the pipeline is located in ./descriptors/analysis_engines/demo.xml. The first part of the pipeline definition is auto-generated during compilation, and points to the meta definition of all components that we are able to run using the current demo.xml. Lines 128-138 are the interesting bit, that define the actual pipeline, that will be executed sequentially.
128 129 130 131 132 133 134 135 136 137 138
<fixedFlow> <node>CollectionReader</node> <node>ImagePreprocessor</node> <node>PointCloudFilter</node> <node>NormalEstimator</node> <node>PlaneAnnotator</node> <node>ImageSegmentationAnnotator</node> <node>PointCloudClusterExtractor</node> <node>ClusterMerger</node> <node>ResultAdvertiser</node> </fixedFlow>
Try to add <node>Cluster3DGeometryAnnotator</node> after the ClusterMerger and relaunch RoboSherlock (no compilation required). You will now have the estimated 3D bounding box with a pose estimate for each cluster (search the visual output for the results).