Abstract:
The visual appearance of an object is not enough to describe an object.A
function is required to complete an object denition.What is the purpose of
an object?How can human possibly interact with an object?
The robot is trained to interact in any environment with no guidance
involved.First the robot learns the interaction from one environment and then
it can transfer its learning to unknown environment.For instance,if a robot
learns to interact with a chair in one environment.Now if the same chair is
found in any of the environment,the robot knows from prior experience that
the chair is used for sitting purpose and hence it performs action around the
detected chair in new environment.
An unsupervised approach is introduced that nds repeating contents
from di erent environment.Human action is observed against single detected
object,so as soon as the similar content is detected in any of the environment
the experience is transferred to all those environments,hence a ordance is
known.
Previously,aordance was determined either by geometry or some supervised algorithms were used.This is rst time an unsupervised algorithm is
introduced for aordance determination.This is a state of the art technique
that de nes new path for research in scene understanding and human-object
interaction in unforeseen circumstances
