5.4. Outdoor-like scenario
In this scenario, the robots move in an outdoor-like rough
environment, which is modeled by a 3D triangular mesh with
2058 triangles. Similarly to the previous scenario, two types of
motion primitives are used to show the effect of motion planning:
(a) primitives learned on the plane environment (RRT-MP/P); and
(b) primitives learned on the same rough surface (RRT-MP/S).
The primitives of RRT-MP/S are learned using the PSO approach
described above, but the fitness function is evaluated on the rough
surface. The algorithms are tested in the same manner as in the
Plane scenario. The test set T is generated to contain M = 100
random start/goal pairs. Each planner is executed for 30 trials with
each start/goal pair. The values are averaged over the 30 trials and
shown as boxplots in Fig. 11. The examples of found solutions are
depicted in Fig. 12.