Model generation
In order to have a robot model in GPU-Voxels you need to provide pointcloud representations of all robot links together with a description of the robot’s kinematical structure. You can choose between a DH-parameter model and a URDF model.
For the voxelization of meshes we rely on binvox by Patrick Min. It is a great GPU-based rasterizer that scans meshfiles and generates binary voxelmaps. By default the user defines the size of the desired voxelspace and the tool will scale the mesh to fit into the space. To keep the scale and the coordinate systems origin of the model, the user can also specify a boundingbox.
As an example you can see the process for the upper body part of our HoLLiE robot model:
Voxelization of robot models
We supply a script (within the model folder) that will automatically set the boundingbox and the number of voxels for a given mesh file and run it through binvox:
voxelize.sh path_to_mesh_file scaling_factor
It will keep the filenames but change the extension. Just copy the .binvox files to the model folder of your GPU-Voxels directory and you can use them with your URDF or DH robot.
For more information: https://github.com/fzi-forschungszentrum-informatik/gpu-voxels