<?xml version="1.0" encoding="UTF-8"?><xml><records><record><source-app name="Biblio" version="6.x">Drupal-Biblio</source-app><ref-type>47</ref-type><contributors><authors><author><style face="normal" font="default" size="100%">Lallée, Stephane</style></author><author><style face="normal" font="default" size="100%">Felix Warneken</style></author><author><style face="normal" font="default" size="100%">Peter F. Dominey</style></author></authors></contributors><titles><title><style face="normal" font="default" size="100%">Learning to Collaborate by Observation </style></title><secondary-title><style face="normal" font="default" size="100%">IEEE Humanoids Workshop on developmental psychology contributions to cooperative human robot interaction</style></secondary-title></titles><dates><year><style  face="normal" font="default" size="100%">2009</style></year></dates><urls><related-urls><url><style face="normal" font="default" size="100%">http://organic.elis.ugent.be/sites/organic.elis.ugent.be/files/Humanoids2009vLalleWarnekenDominey.pdf</style></url></related-urls></urls><abstract><style face="normal" font="default" size="100%">&lt;p&gt;A significant open challenge in human-robot&lt;br /&gt;
interaction is how to transfer task knowledge from the&lt;br /&gt;
humans to the robot. This is particularly challenging in the&lt;br /&gt;
domain of collaborative interaction in which the robot and&lt;br /&gt;
human should take turns in a structured shared plan.&lt;br /&gt;
Interestingly, human infants display a remarkable capacity&lt;br /&gt;
to learn collaborative behavior from a single demonstration,&lt;br /&gt;
and to use this knowledge to take either agent&amp;rsquo;s role in the&lt;br /&gt;
collaborative behavior.&lt;br /&gt;
They are able to extract&lt;br /&gt;
individual&amp;rsquo;s actions in terms of their object manipulation&lt;br /&gt;
goals and attribute these to the appropriate agent, forming&lt;br /&gt;
a &amp;ldquo;bird&amp;rsquo;s eye view&amp;rdquo; of the collaborative action.&lt;br /&gt;
The current research exploits these concepts to allow the&lt;br /&gt;
iCub humanoid to learn cooperative tasks via single&lt;br /&gt;
observations of human demonstration. The tasks involve&lt;br /&gt;
two agents performing coordinated, collaborative sequences&lt;br /&gt;
of simple object manipulations. Action perception is&lt;br /&gt;
organized around physical properties of objects &amp;ndash; their&lt;br /&gt;
appearance and disappearance. The robot has a pre-&lt;br /&gt;
learned action repertoire that mirrors this perceptual&lt;br /&gt;
capability for actions. During human demonstration our&lt;br /&gt;
real-time action parser extracts the sequence of actions,&lt;br /&gt;
including agent attribution. The human and robot then&lt;br /&gt;
agree on &amp;ldquo;who goes first&amp;rdquo; and the shared plan is used by the&lt;br /&gt;
robot to collaborate, taking the appropriate role in the&lt;br /&gt;
learned action plan. We present results from 2 experiments&lt;br /&gt;
in which distinct collaborative behaviors are learned in&lt;br /&gt;
real-time. We argue that this approach provides a powerful&lt;br /&gt;
complement to existing programming by demonstration&lt;br /&gt;
methods.&lt;/p&gt;</style></abstract></record></records></xml>