Tutorial on Embodiment

4. Information Theoretic Implications of Embodiment

In the previous section, we have demonstrated that remarkable behavior can be achieved through exploitation of the properties of the body and its interaction with the environment - with zero or little control. However, this view overly polarizes the situation. Body and brain should not be viewed as competitors, but rather collaborators. In more complex scenarios, mechanical ‘intelligence' has to be aided by software or control. In order for a controller to be able to take the right decisions and issue proper motor commands, it needs to perceive the relevant information regarding the agent's interaction with the environment. Our goal in this section is to emphasize that embodiment plays a key part in this process as well - by shaping and structuring the information that reaches the brain (or controller).

In the section "Leg coordination in insect walking", we have seen how body dynamics and interaction with the environment can be exploited for communication among local controllers in the insect's legs. In this section, we want to specifically address two components that have a major impact on the amount of information that reaches the brain. They are: (i) morphology of the sensory apparatus, and (ii) active generation of information through sensory-motor coordination. We will again illustrate the concepts on animal and robot case studies. We will also review methods that allow us to quantify the information structure in the perception-action loop. These methods are applicable to biological and artificial agents alike; however, our selection will be biased toward robotic case studies, because these allow us to manipulate the morphology and also provide direct access to the information that reaches a robot's "brain".