ros-industrial/reach

Reach study results heat map

Closed this issue · 5 comments

Integrate the code for creating a heat map on the target mesh using reach study results

Hi, thanks for the great work on this repository! We're using it to quantify some design decision of our new mobile manipulator.

Is there currently a way to create a heatmap similar to the one from the readme ? Or is this not integrated ?

It's currently not integrated into the repository. I had to write a custom C++ radial basis function interpolation library to interpolate scores at each vertex of the mesh and some other code that I never ended up formalizing. More recently, I implemented the heat map visualization in Python using scipy and I was a lot happier with it. I did that work on a client project, so I've asked if I can contribute it back to this repo. I should hear back within the week. I can provide more information on the approach though if you're looking to implement this yourself sooner

Thanks for the response! I realized that for our usecase a heatmap on the mesh itself might not be the optimal representation after all. We're trying to check the reachability of picking items of a shelf, so I manually created a pcd file with points on the shelf.

Screenshot from 2022-11-08 11-03-28

The heatmap visualization on the markers is already quite useful. But I would still be interested in the scipy implementation for future projects, if it's okay to make public! Thanks again.

@c-salmi I was able to get approval to share this script. I've included in #38 as a part of the Python interface that we're introducing for this package.

It usually works well with the default parameters, but adding a small amount of smoothing (e.g., 1.0e-3) sometimes helps. There is also radial basis function cross-validation script in the same PR to help with the selection of the RBF kernel; I've found that the default kernel (thin-plate-spline) almost always performs the best. Here's a picture of the output for the demo in the reach_ros package:

image

This may not be the best tool for your application since there does not appear to be any geometry directly at your reach targets, but it may still be of interest

Closing; addressed in #38; feel free to continue the discussion here though