Solving this miracle can be done with a combination of a story graph, interactive execution of high-level commands and a planner. At first, the robot needs a control interface which accepts high-level commands, e.g. “go to south of u-shape”, “go to inside u-shape”. Then a random sequence of such motion primitives is generated until a valid plan is found.
In the figure the semantic enriched motion plan is shown. At first, the robot has go back to it's root position, the next subgoal is “south of the u-shape” and then the robot can go to the goal. For realizing such a planner a datastructure is needed which can be realized as a class. This datastructure contains the graph. It has fields like “parentid”, “name of motion primitive” and “costs”. Such a datastructure is similar to Probabilistic roadmap (PRM), but is enriched with domain knowledge. Here, domain knowledge is the natural language description for the motion primitives.