Towards Adaptive and Autonomous Robots Developing Visually Guided Object Manipulation on the iCub Humanoid

Although robotics has seen advances over the last decades they are still not in wide-spread use outside industrial applications. Proposed scenarios for robots range from cleaning tasks, grocery shopping to elderly care, helping in a hospital, etc. These involve the robots working together, helping and coexisting with humans in daily life. From this the need to deal with a more unstructured environment arises. Object manipulation, which is of high importance in these scenarios, is still a hard problem in robotics. Humans, in contrast, are able to quickly, without much thought, perform a variety of object manipulation tasks on arbitrary objects. The goal of this research is to overcome the limitations of current robotic object manipulation in unstructured environments. A developmental, step-wise approach is used to generate adaptive and autonomous grasping behaviours for novel objects on the iCub humanoid robot. To facilitate this a combined integration of computer vision, artificial intelligence and machine learning is employed. Research Advisor Research Co-advisor Prof. Jürgen Schmidhuber Dr. Alexander Förster Academic Advisor Review Committee Research Advisor’s approval (Prof. Jürgen Schmidhuber): Date: ............................... PhD Director’s approval (Prof. Antonio Carzaniga): Date: ...............................