Thoughts on getting InformedRRTstar to work

[PhilFreeman]

This is probably not appropriate as an "issue" so if there is a better spot to have this conversation I'm happy to move it over there.

I'm trying to see if I can get InformedRRTstar to work as a planner. This is turning out to be way harder than expected. My robot is a simple 3-joint manipulator - so the resulting generated state space is a compound state space with 3 subspaces, each a 1-D real vector space. This totally confuses the informed sampler (PathLengthDirectInfSampler) which expects a compound space to be only one of SE(3) or SE(2). The sampler expects a joint-space to be passed as a simple real vector state space of the appropriate dimension.

So, how might I hack around this? I really only need it to work for my classroom example so the hack doesn't have to be robust. One thought is to write my own informed sampler based on the existing one and recompile OMPL to use that. However, I'm unfamiliar enough the structure of what's going on there that I'm not sure where to start - but this solution is still somewhat appealing as it seems the most "correct".

Another idea is to create a dummy object in V-rep that can be used as the handle for a position 3D state space and let the planner use that. What I can't figure out how to do is link the position of the dummy object to the joint configuration of the robot so that collision checking works.

The last approach would be to modify the plugin so that the state space is not compound, but a simple real-vectored state space. I would imagine it means adding a new method to allocate an N-dim real vector state space and adding a new method to assign the dimensions to joints. My guess is that I'm overlooking something critical here and the call backs from OMPL to V-rep will collapse miserably.

What's a good way to try and approach this problem - because I'm pretty stuck after having given this some serious thought?