Humans use learned knowledge to solve reaching tasks and to manipulate objects and tools.We believe that representations of manipulation characteristics of an object and of the reaching capabilities of a robotic arm can speed up low-level planners, like grasp planners. They also enable sophisticated scene analysis and reasoning for high-level planners, like task planners.
We present object-specific grasp maps to encapsulate an object’s manipulation characteristics. A grasp planner is shown to use the grasps maps and a representation of the reachable workspace. The exploitation of the provided knowledge focuses the planning on regions of the object that are promising to yield high quality grasps. Speed ups of factor 2-12 are reported.