Note that genmos_object_search is written in Python 3. For ROS, it is best integrated with ROS Noetic.
Go to the `src` folder of your ROS workspace. Then run:
ln -s path/to/genmos_object_search/ros genmos_object_search_ros
This effectively adds a ROS package called “genmos_object_search_ros” into your workspace
This is a ROS package; Therefore, it is expected to operate within a ROS workspace.
Before building this package, make sure you have activated a virtualenv. Then, run
source install_dependencies.bash
to install python dependencies.
Note that we use ros_numpy to convert point cloud ROS message to numpy array,
so that point cloud observation can be used to update the search region (see code).
Typically, one installs ros_numpy by sudo apt-get install ros-noetic-ros-numpy.
However, there are two issues (1) this issue where numpy > 1.24 causes an AttributeError on `numpy.float`,
and that (2) the ros_numpy package was no longer maintained.
As a result, we folded the source code of ros_numpy into our repository,
under ./thirdparty/ros_numpy. It is modified to be a standard Python
package that can be pip-installed. The installation is, still, handled
by source install_dependencies.bash so you do not need to do anything.
catkin_make -DCATKIN_WHITELIST_PACKAGES="genmos_object_search_ros"
Note that if you are using the [robotdev Spot environment](https://github.com/zkytony/robotdev/tree/master/spot), you need to run the following command instead
build_spot -DCATKIN_WHITELIST_PACKAGES="genmos_object_search_ros"
Install gdown, then run download script:
pip install gdown python download.py
This will download both the SL\_OSM\_Dataset and the frame of reference
prediction models. The SL\_OSM\_Dataset will be saved under data, while the
models will be saved under models.
Also, you will need to download spacy models. We use en_core_web_lg (400MB). To download it:
python -m spacy download en_core_web_lg