Abstract:
Intuitive learning of new behaviours is one of the important aspects of social robotics. Among various robot learning approaches, recently Programming by Demonstration (PbD) has gained significant recognition with a lot of potential. Internal representation of the knowledge is a key design choice in the learning process. Using machine learning techniques such as ANNs, HMMs and NARMAX models, simple skills can be encoded from raw sensory data. However, the abstract symbolic representations have demonstrated greater potential for learning complicated tasks but with less details and require a piece of prior knowledge as well. For a particular application, appropriate choice of the symbols is a key design issue. This paper discusses the choice of the symbols to build a PbD process for typical indoor navigation. The learning results are presented for a few tasks to demonstrate the potential of the proposed approach.