Task Planning Based on the Interpretation of Spatial Structures

Original scientific paper In this research, a new task planning algorithm is developed for building a desired object configuration from a given initial unordered object state. The task of the planning algorithm is to find a feasible set of actions, i.e. a finite number of discrete transformations, which can rearrange the objects into a desired ordered final state. The environment is interpreted through the position and orientation of the objects. The solution to the planning problem is proposed as a two-step method. First, a constructive heuristic generates an initial set of good solutions. The constructive heuristic uses only mutations for making an initial population of state transitions. A genetic algorithm is developed for optimizing the initial set of solutions. The genetic algorithm is characterized by a parallel evolutionary strategy, with the aim of spatial transformation of unordered object states into ordered object states. The algorithm can be used for solving the task planning problems represented in the two-dimensional space. Verification of the planning algorithm is done in a virtual environment.