More than two trees

If a dual-tree approach offers advantages over a single tree, then it is natural to ask whether growing three or more RDTs might be even better. This is particularly helpful for problems like the double bug trap in Figure 5.13c. New trees can be grown from parts of ${\cal C}$ that are difficult to reach. Controlling the number of trees and determining when to attempt connections between them is difficult. Some interesting recent work has been done in this direction [82,918,919].

These additional trees could be started at arbitrary (possibly random) configurations. As more trees are considered, a complicated decision problem arises. The computation time must be divided between attempting to explore the space and attempting to connect trees to each other. It is also not clear which connections should be attempted. Many research issues remain in the development of this and other RRT-based planners. A limiting case would be to start a new tree from every sample in ${\alpha }(i)$ and to try to connect nearby trees whenever possible. This approach results in a graph that covers the space in a nice way that is independent of the query. This leads to the main topic of the next section.