Grasp affordances from multi-fingered tactile exploration using dynamic potential fields

In this paper, we address the problem of tactile exploration and subsequent extraction of grasp hypotheses for unknown objects with a multi-fingered anthropomorphic robot hand. We present extensions on our tactile exploration strategy for unknown objects based on a dynamic potential field approach resulting in selective exploration in regions of interest. In the subsequent feature extraction, faces found in the object model are considered to generate grasp affordances. Candidate grasps are validated in a four stage filtering pipeline to eliminate impossible grasps. To evaluate our approach, experiments were carried out in a detailed physics simulation using models of the five-finger hand and the test objects.