The robot perception package is used by Voxeland to provide the input point clouds. Specifically, it is responsible for synchronizing the color and depth images, (optionally) interfacing with the semantic segmentation provider, and generating a 3D point cloud from the resulting information.
See the launch files for the parameters.
To use without semantic information (and, thus, without depending on an external classifier), select pointcloud_type which does not contain semantic information (XYZ or RGB).
To use WITH semantics, see the readme of the main Voxeland repo, which links to compatible semantic segmentation packages.
The requirements file included in this repo lists the (non-ROS) external dependencies. The ROS packages that this project needs are listed in the standard way, through the package.xml file, so that they can be installed with rosdep.