Hi I would like to test a motion planning algorithm for dynamic environments based on point cloud. I have a dataset of recorded point cloud from a kinect.

I was wondering how to perform collision detection between the robot and the pointcloud and in particular how to avoid compenetration.

https://robotics.stackexchange.com/questions/8546/modelling-point-clouds-for-collision-detection-in-gazebo

Here is shown that i can use PCL to convert the point cloud in a STL mesh, but can can I do it online and "publish" in gazebo a new mesh at a freq of 30Hz? If yes I suppose that the simulation might became extremely slow due to convertion (point cloud -> STL), right?. In that case I could make a dataset of meshes each one corresponding to each frame recorded.

Does exist another method?