Perception through Anticipation –- An Approach to Behaviour-based Perception

lations are provided by the 'reaction block'. This block can be pre-wired or trained. The appraisal block will usually be prewired and associates values to certain sensory aspects. Figure 15 visualizes the cooperation of the three blocks. Starting from the given situation, the reaction block proposes a motor command that should be simulated first. With the sensory situation and this motor command , the anticipation block predicts the next situation, which is in turn evaluated by the appraisal system. This process is continued for a number of steps. The output of the appraisal system for all situations in the sequence is combined to a cumulative value. If this value indicates a successful sequence, the first motor command will be executed ; if the sequence is assigned a negative value, another sequence will be generated (e.g. by using the second best proposal of the reaction block). The better the proposals of the reaction block, the more the system will appear to be 'reactive' seen from the outside. This will usually be the case for situations that are not too complex and that are, therefore, completely covered by the latent reactions. In complex situations (like the dead end situations the agent in section 'Reaction vs. Perception vs. Anticipation' has to cope with), the anticipation block will come in, and the selected sequence will usually not be the one generated first. Summary An approach to behaviour-based perception was presented that tries to overcome the problems of the information processing metaphor. Perception of space and shape is assumed to be a generative process of anticipating the course of events resulting from different sequences of actions. That way, perception and generation of behaviour can be fused into a single neural process. As shown for a simple example, this approach could provide another way to cope with complex recognition problems, including invariance and constancy problems. Perception of space and shape is considered as a basic cognitive ability, arising from an extension of reactive systems and providing the basis for higher abilities like functional categori-zation. Two ways of modeling were examined: a microscopic model close to neurobiology, that is based on a modified assembly theory, and a macroscopic model realized at a coarse scale. Whereas with the microscopic model it could be shown that the process of perception and generation of behaviour can be projected onto a real nervous system (contributing a functional interpretation of the cortical …