Learning Tasks and Skills Together From a Human Teacher

Introduction Robot Learning from Demonstration (LfD) research deals with the challenges of enabling humans to teach robots novel skills and tasks (Argall et al. 2009). The practical importance of LfD is due to the fact that it is impossible to pre-program all the necessary skills and task knowledge that a robot might need during its life-cycle. This poses many interesting application areas for LfD ranging from houses to factory floors. An important motivation for our research agenda is that in many of the practical LfD applications, the teacher will be an everyday end-user, not an expert in Machine Learning or robotics. Thus, our research explores the ways in which Machine Learning can exploit human social learning interactions—Socially Guided Machine Learning (SG-ML). We claim computational learning systems should make use of social learning principles because: 1. It is better for the human: A working hypothesis of this approach is that using characteristic aspects of human learning will be the most natural interface for everyday people. Several studies show that humans inherently and dynamically provide social scaffolding for learners. Greenfield describes studies of children learning language and girls learning to weave (Greenfield 1984), where the teachers engage in a complex process of dynamically adjusting their support of the learner based on skill level and success. The teachers are unconscious of the process or the method by which they are teaching. Thus, the partnership of social learning is an interaction that people fall into intuitively. 2. It is better for the machine: This point is generally less intuitive, but one way to think of it is that social interaction provides biases and constraints that simplify the problem for the machine. Social learning leads to a more efficient and robust machine learning process. We have several examples of this in our prior work (Thomaz and Breazeal 2008; Thomaz and Cakmak 2009; Cakmak et al. 2010; Zang et al. 2010). We are interested both in novel algorithms and systems for LfD and in using everyday people as our benchmark for evaluation. As part of the AAAI 2011 LfD Challenge, we will be exhibiting our SG-ML approach to LfD.