CoSTAR does not rely on any particular perception solution. Instead, we require a couple specific outputs, including TF frames and object detection messages.

This is our initial perception package. See here for more information.