Seems to be rather alot of parallelisation happening these days.
I understand that it takes time to collaborate and that natural selection from multiple implementations are also a way of evolving the code available in our robotic community, but I do believe at least some preliminary discussion and collaboration on software that 90% fits can often get you that last 10% much faster than evolving it entirely yourself. It also rounds out an implementation to be more satisfactory for everyone. Another problem everpresent with ROS software is maintenance - n implementations are of no use long term if maintainers can't keep up with all that they try to maintain.
I'd like to encourage people to first look to see what is there, see if something is a 90% fit, discuss and then iterate using the usual fork/pull request cycle where possible.