The above-mentioned locomotion generators as well as motion
planning methods can provide motion for modular robots.
Locomotion generators are efficient in providing basic motions,
but they do not consider complex situations in the environment.
Although generators can be extended to cope with altering terrain,
finding suitable parameters for such models is time consuming.
However, motion planning techniques are well fitted to solve the
global planning problem, as they cope with obstacles and consider
the situation on a higher level. However, their performance can
degrade when they need to derive control signals for many
actuators, which is the case of modular robots.
In this paper, we propose the novel RRT-MP planner (Rapidly
Exploring Random Tree with Motion Primitives), that combines
locomotion generation and motion planning. To fulfill a highlevel
goal, sampling-based motion planning is utilized, as it
considers the overall situation in the environment and it can
avoid collisions with obstacles or other robots. To speed up the
planning process, the problem of finding control inputs is solved
by utilizing locomotion generators that provide basic skills—
motion primitives. The result of RRT-MP is a sequence of motion
primitives, which can easily be executed by the robot. Whenever
the situation in the environment changes, a new plan can easily
be generated without the necessity to adapt the individual motion
primitives.
The main contribution of our approach is the novel motion
planner, which can create plans for modular robots moving in
complex environments. Our approach can be combined with
various locomotion generators, which can realize different gaits.
The motion planner allows the use of simple motion primitives
without the need to introduce sensory feedback to detect obstacles
or other difficult areas. The proposed system allows the robot
to visit distant places in complex environments, where obstacles
need to be avoided. It is assumed, that the dynamics of the robots is
not crucial here, as the studied robots move mostly using crawling
and therefore the impact of dynamics is negligible and it can be
handled on the CPG level. This allows us to focus this paper only to
the motion planning problem.