The goal in the local solution is to achieve the desired position of one end effector, whereas in the global solution, the goal is to produce such a sequence of motions, so that the whole model can reach the desired position and orientation. The first approach is directly connected to solving inverse kinematics.
That means, the genetic programming based motion control could be applied not only to the level of partial movements, but it could be also used to control high level motions (like step, jump, etc.) [Srk99]. The motion simulation is controlled with partial abstraction and by goals for achievement.
There are conditions and limits constructed for each certain task and the evaluation function of success. A number of solutions are combined in each generation and the best solutions according to evaluation function are chosen. The next generation arises by crossing (hybridization) and mutating of best solutions from previous generation. The generations repeat until desired result is achieved.