Constrained Deformable Coherent Point Drift (CDCPD)
CDCPD is a implementation of Occlusion-robust Deformable Object Tracking without Physics Simulation by Cheng Chi and Dmitry Berenson.
This library includes object-oriented API for tracker, visualization utilities and a ROS node that subscribes and publishes PointCloud2.
Requirements
*: (optional) for visualization only
Installation
Note: these instructions have been tested on Ubuntu 18.04 and ROS Melodic only.
Core Package
Clone the following repositories into your catkin workspace:
cd path/to/catkin_ws/src/
git clone git@github.com:UM-ARM-Lab/cdcpd.git
git clone git@github.com:eric-wieser/ros_numpy.git
Build the project (tested with catkin build, not catkin_make)
catkin build
Even if you don't use this, you should still do these steps because the code imports libfeatures. If you need tracking failure recovery, you will also need to compile the VFH feature library. Make sure you have PCL-1.8 installed. If you have ROS installed, it should be already there.
cd cdcpd/src/pcl_features
mkdir build
cd build
cmake ..
make
cd ..
ln -s build/libfeatures.so
Conda configuration
Install miniconda https://docs.conda.io/en/latest/miniconda.html; use the python 3 version
Create conda virtual enviornment with the packages that conda knows about, you must name it as shown for the roslaunch/scripts to run correctly. This environment is used to isolate python3 code from the default python2 code on Ubuntu 18.04 and ROS Melodic
conda create -c gurobi --name catkin_cdcpd python=3.7 opencv gurobi numpy scipy numexpr scikit-learn matplotlib ipython
conda activate catkin_cdcpd
For the following operations, make sure the newly created virtual environment is activated. The command line should show the name of virtual environment in brackets. Install the rest of the requirements via pip:
(catkin_cdcpd) user@host:path/to/catkin_ws/src/cdcpd$ pip install -r cdcpd/requirements.txt
Gurobi Licence
Gurobi is a proprietary optimization package that we use. Please obtain a free academic license. Note that Gurobi licence activation requires a university network. If you are not on campus, follow these instructions to setup a VPN: UMVPN.
Ensure that the conda environment is active, and install a Gurobi key with grbgetkey following the instructions here
(catkin_cdcpd) user@host:~$ grbgetkey 253e22f3-...
You may have noticed that a package named glplotlib is installed by pip. glplotlib is another opensource library developed by Cheng Chi for visualizing 3D point cloud with better performance (compared to matplotlib).
If you use a IDE such as PyCharm or VS Code, remember to set project interpreter to this virtual environment.
Demo
To run the demo, you will need to download some data into <project_root>/data folder.
The files in the v0.1.0 folder are intended for use with the original version of this library. Follow instructions in the README there to use them.
To run the fully ROS integrated demo start all of the following from individual terminals, without activating the conda workspace:
- Start ROS core:
roscore - Start Rviz for visualization:
rviz -d path/to/catkin_ws/src/cdcpd/config/cdcpd.rviz - Start the tracking nodes:
roslaunch cdcpd tracker.launch - Start playing the rosbag:
rosbag play -l data/placemat_stationary.bag
The following messages are a normal part of the initialization process:
[rosout]: Error: itemsize cannot be zero in type
/path/to/catkin_ws/src/cdcpd/src/cdcpd/cv_utils.py:68: RuntimeWarning: divide by zero encountered in true_divide
projected[:, 0] /= projected[:, 2]
/path/to/catkin_ws/src/cdcpd/src/cdcpd/cv_utils.py:68: RuntimeWarning: invalid value encountered in true_divide
projected[:, 0] /= projected[:, 2]
/path/to/catkin_ws/src/cdcpd/src/cdcpd/cv_utils.py:69: RuntimeWarning: divide by zero encountered in true_divide
projected[:, 1] /= projected[:, 2]
/path/to/catkin_ws/src/cdcpd/src/cdcpd/cv_utils.py:69: RuntimeWarning: invalid value encountered in true_divide
projected[:, 1] /= projected[:, 2]