Planning failures while using Persistent PRM roadmap

[stephanie-eng]

Description

I tried to use PersistentPRM with a project and noticed no planning time improvements. When trying to replicate this issue with a less complex setup, I found expected behaviour with the Panda, but some possibly unexpected behaviour when I added a more complex collision environment (the same collision environment from #492).

I tried to plan close to a collision object (see below), and successfully planned with PersistentPRMStar and saved a roadmap. I tried the same plan again with PersistentPRM and loading the roadmap, and I sometimes got planning failures. Since I'm using a cached roadmap, I'd expect that I wouldn't see planning failures.

I've created a branch of moveit2_tutorials with the collision environment from #492 for testing here and a branch of moveit_resources with my ompl_planning.yaml using roadmap caching here.

[tylerjw]

How can we improve the tutorial site to make this better?