Using semantic information for improving efficiency of robot task planning

The use of semantic information in robotics is an emergent field of research. As a supplement to other types of information, like geometrical or topological, semantics can improve mobile robot reasoning or knowledge inference, and can also facilitate human-robot communication. These abilities are particularly relevant for robots intended to operate in everyday environments populated by people, which typically involve a great number of objects, places, and possible actions. In this paper, we explore a novel usage of semantic information: as an improvement for task planning in complex scenarios like the mentioned ones, where other planners easily find intractable situations. More specifically, we propose to first construct a "semantic" plan composed of categories of objects, places, etc. that solves a "generalized" version of the requested task, and then to use that plan for discarding irrelevant information in the definitive planning carried out on the symbolic instances of those elements (that correspond to physical elements of the world with which the robot can operate). Our results using this approach are promising, and have been compared to other existing approaches.