About

Interaction is the ultimate goal of humanoid robots, therefore social behavior and social
cues are essential in particular for learning. In recent years insight has grown that
learning heavily draws on social cues provided by the interaction in which the tutoring
takes place. For example, it has been shown that parents modify not only their speech
but also their movements when teaching their infants new objects and actions. Similarly,
it has been shown that these and other modifications can also be observed in humanrobot
interaction. This is supported by findings that suggest that the more human‐like a
robot looks, the more humans speculate about its intentions and reasoning. Already
there exist first approaches to automatically detect these changes in behavior and make
use of them, e.g. in order to segment actions into smaller units. However, it is still
unclear if and how exactly these cues may help learning.

On the other hand, in imitation learning the focus is still often on settings that ignore
such situational influences, except for very few approaches that take e.g. verbal
emphasis into account or emphasize the importance of social learning. Many classical
approaches try to model mean trajectories whereas in developmental research it is
currently being discussed how the high variability of hand trajectories observed in
infant tutoring may help infants to learn. While imitation learning has for a long time
been focussing on learning trajectories only, there now exists work on how to
automatically recognize the desired effect of an action.

In this workshop we want to bring together researchers from the field of robotics with
experts from developmental learning in order to discuss issues of learning from both
perspectives. We therefore invite papers from the field of imitation learning as well as
from developmental learning with a focus on the question of how actions and behavior
can be learned. Further important topics in the vein are studies on the shaping, impact
and measurement of social interaction between robots and human, and the enabling role
such social interactions can have on learning of basic perceptual or motor skills.

Target questions:

• Is an understanding of the action / behavior necessary to learn it?
• What additional information is given in tutoring situations and how can it help to learn actions / behavior?
• How far can current imitation learning approaches reach?
• What does physical interaction between tutor and robot provide?
• How to signal / recognize what and when to learn in interaction?
• How to shape a tutoring interaction with a humanoid robot?
• How does interaction help to segment actions perceptually?