Basic Functional Modeling

First we define Active Vision Agent (AVA, in short) as a rational agent with visual perception, action, and communication capabilities. Let denote the internal state of ith AVA, and the state of the world where a group of AVAs are embeded.

Then, intuitively, perception and action by a single can be modeled by the following state mapping functions (Fig. 4):

  

where denotes entities perceived by . We introduce to link and . It denotes the perception-driven and/or autonomous internal state transition by :

 

  
Figure 4: A primitive model of interaction between perception and action.

Since includes various camera parameters such as camera position, viewing angle, zooming factor, focus, and so on, depends on ; changes depending on . , on the other hand, can modify its own internal state as well as the state of the world :

• AVA with a physically actionable body (e.g. one implemented on a mobile robot with manipulators) can change both and . We call such AVA embodied AVA. Equation (2) models such physical actions. Note that this model is a naive one to be refined later.
• When AVA is equipped only with active camera(s), on the other hand, its action implies the change of its internal state without any side effects on the world state. We call such AVA vacuous AVA and model its action by

   

  We call the implementation of a vacuous AVA an Observation Station: real time image processor with active camera(s) embedded in the world.

The discrimination between embodied and vacuous AVAs plays a crucial role in defining the meaning of communication.