RRTs first published in  are randomised planners especially adept at solving difficult,high-dimensional path planning problems. However, environments with low-connectivity due to the presence of obstacles can severely affect convergence. Multiple RRTs have been proposed as a means of addressing this issue, however, this approach can adversely affect computational efficiency.
This paper  published by the authors of this Matlab code is the implementation of multiple Rapidly-exploring Random Tree (RRT) algorithm work. This paper introduces a new and simple method which takes advantage of the benefits of multiple trees, whilst ensuring the computational burden of maintaining them is minimised. Results indicate that multiple RRTs are able to reduce the logarithmic complexity of the search, most notably in environments with high obstacle densities.
 LaValle, S. M., ‘Rapidly-Exploring Random Trees: A New Tool for Path Planning’, TR 98-11, Computer Science Department, Iowa State University, Oct. 1998.
 Matthew Clifton, Gavin Paul, Ngai Kwok, Dikai Liu, Da-Long Wang, "Evaluating Performance of Multiple RRTs", IEEE conference on Mechatronic and Embedded Systems and Application, 2008
Dear Gavin, thanks for providing this code. It would be easier to understand and implement if you would add a complete description of all arguments in the header of each function - the same way Matlab does.
There are also name collisions, 'connect' is just too generic. A good workaround is to use a prefix such as 'rrt_' to make function names more unique and to simplify renaming.