zkingston
commented
1 month ago Thank you for the kind words! Glad you find the project useful.
The process for generating the 2 level sphere hierarchy used in the VAMP collision checker is as follows:
- Preprocess the robot's mesh collision geometry to confirm that the meshes are watertight manifolds. We used https://github.com/hjwdzh/Manifold and https://github.com/hjwdzh/ManifoldPlus
- Optionally, simplify the mesh geometry with something like quadric mesh simplification: https://github.com/sp4cerat/Fast-Quadric-Mesh-Simplification
- Use the medial-axis sphere tree generation script from https://github.com/mlund/spheretree to create a two level sphere tree, with desired branching factor. Choosing the desired branching factor should be done to the level of approximation desired - more spheres better approximate the mesh, but increase complexity. This will create two sets of spheres, a large bounding sphere for the entire mesh and a more refined approximation. It is possible to do more levels of hierarchy, see the original sphere tree paper. A lot of the parameters for the sphere tree script are hyperparameters, I believe defaults should work OK. Branch and depth are the main parameters to modify to control sphere tree size and depth.
Also, please subscribe to the issue that is tracking the release of the tracing compiler: https://github.com/KavrakiLab/vamp/issues/7 Part of this release will include a tool to automatically create the spherized URDF of the robot.
ZhouYixuanRobtic
Hello, First of all, I want to thank you for your fantastic work on this project. It has been incredibly valuable, and I really appreciate the effort and expertise you've put into it. I would like to request a more detailed explanation of the sphere-tree construction process in the documentation or code comments. While reviewing the current materials, I found that some steps in the construction process were unclear, especially regarding how the spheres are generated, combined, and their hierarchy structured.
It would be extremely helpful if the following aspects could be explained in more detail:
The algorithm used for constructing the sphere-tree. I found that https://github.com/mlund/spheretree takes vague description. And hard to combine them into a sphere-based robot. The criteria for sphere combination at different levels. Any optimization strategies applied during the construction. This clarification would greatly benefit developers trying to understand or modify the implementation.
Thank you for considering this request!