Introduction
An affordance, a concept from Ecological Psychology, denotes a specific relationship
between an animal and its environment. Perceiving an affordance means perceiving an
interaction possibility that is specific for the
animal's perception and action
capabilities (Fig. 1). The MACS project investigated how this concept can be exploited to
improve a technical system, a mobile robot with manipulation capabilities.
The perception capabilities of mobile robots comprise detection and recognition
algorithms that work on sensory data. The richest data are usually provided by
camera images, and most employed algorithms stem from the field of Computer Vision.
Appearance-based object recognition is a widely employed method for robot
perception. But given that reliable recognition of everyday things is currently
restricted to low numbers of objects, such methods seem too restrictive for
reliable robot performance in everyday environments, particularly when mobile
manipulation is involved.
But what are the alternatives? Starting from the consideration that it is most
important that a robot will get its job done, it is clear that a robot would
benefit from abilities to find alternative solutions to a given task. The role
model is the human ability to improvise, e.g. to use artefacts in ways that they
were not designed for or to use one object instead of another one, provided they
offer similar functionalities. A human could easily decide to use a mug as a drinking
vessel if a glass is not available. Glasses and mugs have quite different appearances.
But what is important in this example is their function as a drinking vessel, not the
particular appearance. Fig. 2 shows another example,
namely opportunities for a human to sit.
Transferred to a technical system, perceiving an affordance does not include
appearance-based object recognition, but rather feature-based perception of
(object) functions. The central hypothesis of MACS was that an affordance-inspired
control architecture enables a robot to perceive more interaction possibilities
than a traditional architecture that relies on appearance-based object recognition
alone.
This project web site offers you the results of
the MACS project, including the list of project
publications and public deliverables, a link to an extensive online
bibliography on affordance-related research, videos
from the experiments, a list of events that we
organized or attended, and more.
|