Decimate Collada/urdf for faster collision checking in MoveIt

Question

Hey guys,

I got a couple of .dae files of my robot in fairly high detail. After creating the .urdf with all the joint limits and stuff, I have created a MoveIt-package with those files and can now see the robot and plan in Rviz.

Since my dae files are fairly high detail, I wanted to know if there is a "ROS"-way of decreasing the polynom count in a safe way, so that MoveIt can check faster for collisions.

I didn't create the dae files and don't know much about this format, but my guess would be if I just open them up in a program like blender or something and save them again with less detail, there is no way to guarantee that I didn't accidently remove parts from the robot, letting MoveIt believe that this certain space is empty? Also, some kind of setting for the error would be nice, like "Create me a model which is at maximum 15 mm away from the actual robot"

Is there a script/package/program, that does this?

Thanks in advance, Rabe

---

Originally posted by Rabe on ROS Answers with karma: 683 on 2014-05-26

Post score: 2

Answer 1

Depending on the fidelity you want to achieve and the link shapes of your robot, this might be hard to completely automate. I´d recommend giving either Blender or Meshlab a shot. Both allow generating convex hulls for meshes very easily. These generally have a much lower poly count than the original meshes and you can easily simplify them some more after convexification using Meshlab´s simplification filters. Using convex hull, you´ll always overestimate the colliding volume, which gives you a safety margin (but also can result in a reduction of collision-free workspace size if the convex hulls add too much volume). The arms and legs of this 3D printable Atlas model are an example of using convex hulls of the original (much more complicated) geometry.

The other viable option is specifying the collision geometry manually using primitives such as boxes and cylinders, but that is a much more involved manual process for most practical robot systems.

That´s the two practical approaches I used so far at least, but I´m also interested in other ideas, so keep the answers coming ;)

---

Originally posted by Stefan Kohlbrecher with karma: 24361 on 2014-05-26

This answer was ACCEPTED on the original site

Post score: 3

---

Original comments

Comment by Rabe on 2014-05-26:
Thanks, I created a convex hull with meshlab and this reduced the planning time from 2-4 seconds to under .5 seconds. I am still keeping the question open, since I am also intrested in more automated and use-case-specific ways to do this.