Is Machine Learning Enough to Train Robotic Pets?

We discuss the problem of learning in robotic pets asking whether the core machine learning paradigm, namely the optimisation of a bounded error function, is sufficient in this context. In pet robots, it seems that the learning process itself rather than the result of this process is the main criterion for the quality of the interaction. Potential extensions of the optimisation paradigm include emotional, self-organising, and exploratory mechanisms to support desirable learning capabilities of a robotic pet. We also propose a co-design process that develops a personalised interaction experience and mutual learning with active contributions from both robotic pets and
their owners.