Object Learning through Interactive Manipulation and Foveated Vision

Autonomous robots that operate in unstructured environments must be able to seamlessly expand their knowledge base. To identify and manipulate previously unknown objects, a robot should continuously acquire new object knowledge even when no prior information about the objects or the environment is available. In this paper we propose to improve visual object learning and recognition by exploiting the advantages of foveated vision. The proposed approach first creates object hypotheses in peripheral stereo cameras. Next the robot directs its view towards one of the hypotheses to acquire images of the hypothetical object by foveal cameras. This enables a more thorough investigation of a smaller area of the scene, which is seen in higher resolution. Additional information that is needed to verify the hypothesis comes through interactive manipulation. A teacher or the robot itself induces a change in the scene by manipulating the hypothetical object. We compare two methods for validating the hypotheses in the foveal view and experimentally show the advantage of foveated vision compared to standard active stereo vision that relies on camera systems with a fixed field of view.