Human Machine Interaction and Communication in Cooperative Actions - Bioinspired Approaches for Human-Centric Technologies - page 243

Fig. 8.1 iCub is interacting
with a human partner by
playing together with a
touch screen. The way the
robot moves its arm and its
gaze has a strong influence
on human perception
(Sciutti et al. 2013a ). Photo
by Laura Taverna

8.2.2 Mutual Adaption to Robot Behavior

Another important property of human-human interaction is mutual adaptation.
When we walk or work with someone else, we automatically align the speed of
our actions to our partners' rhythm. Previous studies have shown that in case of
virtual stimuli (e.g., a single dot moving on a screen, ideally representing the finger
of an agent), such motor contagion is selectively present only if the motion follows
a biologically plausible kinematics (Bisio et al. 2010 ). We have evaluated whether a
similar phenomenon extended also to the interaction with a physical, nonhuman
agent, as the robot iCub. In particular, a demonstrator (either human or robot)
performed a reaching or a transport action at different speeds. After action obser-
vation, subjects were instructed to reach the same target or transport the same
object. Motor contagion was assessed by evaluating the influence of the observed
speed on the velocity adopted by subjects during their actions. Different parameters
of the demonstrator's action were manipulated, including the shape of the trajectory
and the kinematics of the motion (biologically plausible or not). Our results
suggested that the nature of the agent may have relevance in determining human
implicit adaptation, but only as a function of the adopted kinematics (Bisio
et al. submitted).

8.2.3 Inferring Object's Properties from Robot Observation

Importantly, through action observation, humans can automatically infer not only
others' goals but also the dynamics properties of the objects manipulated by the
action partner. For instance, humans are extremely precise at estimating the weight
of an object just by observing the lifting action performed by a human partner
(Runeson and Frykholm 1981 ). Moreover, they can also use this information to