This library provides implementations for single and multi-object instance localization from RGB-D sensor (MS Kinect, ASUS Xtion etc.) data. These are based on the PERCH (Perception via Search) and D2P (Discriminatively-guided Deliberative Perception) algorithms.
- Ubuntu 14.04+
- ROS Hydro+ (active development only on Indigo)
- Get ROS Indigo from http://wiki.ros.org/indigo/Installation/Ubuntu
- Set up a catkin workspace ~/my_workspace (http://wiki.ros.org/catkin/Tutorials/create_a_workspace).
- Download the rosinstall file to your workspace.
cd ~/my_workspace
wstool init src
wstool merge -t src perch.rosinstall
wstool update -t src
rosdep install --from-paths src --ignore-src --rosdistro indigo -y
catkin_make -DCMAKE_BUILD_TYPE=ReleaseFirst, download the object CAD models (92 MB total):
roscd sbpl_perception
chmod +x data/scripts/download_demo_models.sh
./data/scripts/download_demo_models.shAn example RGB-D scene containing 3 objects is provided under sbpl_perception/demo. To run PERCH on this with default parameters:
roscd sbpl_perception && mkdir visualization
roslaunch sbpl_perception demo.launch The states 'expanded' as well as the goal state will be saved under sbpl_perception/visualization. The expanded states will also show up on an OpenCV window named "Expansions". To also save all the 'generated' (rendered) states to sbpl_perception/visualization, use
roslaunch sbpl_perception demo.launch image_debug:=trueYou should see the following input depth and output depth images under sbpl_perception/visualization:
Configuration parameters for the algorithm can be found under sbpl_perception/config/demo_env_config.yaml and sbpl_perception/config/demo_planner_config.yaml, along with descriptions of those parameters.
To pull changes from all related repositories in one go, run wstool update -t src.
For more information on setting up PERCH for your custom robot/sensor, reproducing experimental results, and API details, refer to the wiki.