Unified approach to qualitative motion planning in dynamic environments

Traditional motion planning methods rely on precise kinematic models to either compute the goal trajectory off-line, or to make on-line decisions based on current observations from a dynamic environment. With the increasing use of qualitative modeling in cognitive robotics, different planning approaches are needed to handle the lack of numerical data in manually constructed or autonomously learned qualitative domain theories. We propose a new motion planning algorithm that makes on-line decisions based on given qualitative domain description to reach a goal state. Decisions are stated in the form of simple qualitative actions that can easily be interpreted by robot’s controller and transformed to a numerical output. We demonstrate its use on three classical problems: pursuing, obstacle avoidance and object pushing.