The MACS project aims at investigating the potential benefits of affordances as the underlying principle of a complete robot control architecture.
The term affordance was coined by ecological psychologist James Jerome Gibson who used it to describe a function-based view on perception. In this context, an affordance refers to the potentiality for action that an environmental object might afford to a particular agent. A flat, rigid, knee-high, and stable surface, for instance, would afford for a humanoid agent to be something to sit on; in other words: the agent perceives the affordance of sitability as a relation between the just mentioned environmental features and the agent’s own capabilities and constraints.
In contrast to the currently more wide-spread object-based perception approaches, the affordance-based approach does not aim at recognizing objects and to classify them into predefined categories in order to determine what interaction possibilities the objects afford. Instead, an agent directly perceives the functional characteristics of environmental features. It therefore becomes able to, for instance, sit on every appropriate surface rather than merely on those objects it recognizes to be chairs. Moreover, the same gymnastic ball that affords sitting on, may afford tossing if the agent simply looks for something to play with. The affordances of environmental objects which an agent perceives are thus context-dependent and related to the agent itself and its capabilities for action.
To evaluate this theory, a hybrid robot control architecture is being developed that explicitly bases all its different components on the affordance concept making them 1st-class citizens. Within the MACS consortium it is believed that the resulting system will show to benefit of this integration in terms of robustness and generality.
MACS at the Knowledge-Based Systems Group
Within the project consortium, the Knowledge-Based Systems Group of the University of Osnabrück is dealing mainly with the following topics:
- Affordance-based Planning
An affordance-based planning module will be developed as a part of the MACS hybrid robot control architecture. Therefore, a domain description for the MACS specific demonstrator scenarios will be developed yielding, amongst others, a set of affordance-related and normal operators. The system will apply a backward-chaining planner working on a newly developed, unique affordance representation triple.
- Affordance Representation Triple
The MACS hybrid robot control architecture is based on an affordance representation that consists of a cue descriptor, behavior descriptor, and an outcome descriptor. As the planning module will work based on these representations, the three different parts will be specified in detail and will eventually be used in a prototype implementation of the module.
- Execution Control
The system will provide plans consisting of sequences of actions and sets of equivalent affordances to look for that are assumed to lead to the desired goal. Therefore, the actual interfacing between the planning module and the execution control as well as between the execution control and the task execution component of the system have to be specified, developed, and integrated.