next up previous contents
Next: From Perception to Behavior Up: The Focusser Previous: Motor Preferences



To summarize, the focusser delivers to the appropriate behavior component the accurate sensory information about the object of interest as well as the motor preferences reflecting the current environmental conditions. This design not only provides highly focussed attention for efficiency, but also allows influence from environmental conditions to achieve compromised actions.

As is indicated by the examples in the previous section, motor preferences are used in two ways. In the first way they are used to influence the choice of a proper action. This is done by influencing both the selection of a proper motor skill (e.g., example 1 above) and the calculation of the corresponding control parameters. In the case where all possible actions are accompanied by equal-strength negative motor preferences, an action gets chosen at random and the corresponding control parameters (angle or speed) are modified to accommodate the preference. For example, suppose the right-turn-MC is chosen and the calculated angle of turn is tex2html_wrap_inline3300 , since tex2html_wrap_inline3264 tex2html_wrap_inline3304 , the final, compromised angle of turn will be tex2html_wrap_inline3306 . Similar modification of motor control parameters is also performed in the case where the chosen action is accompanied by a negative preference. Positive preferences do not have any effect on the control parameters.

In the second way, the motor preferences are used to help the focusser locate the object of interest. This process demonstrates a certain level of compromise among different desires. Typical cases include targeting the most desirable food source and mating partner. In each case, there is an evaluation criterion according to which a real valued `desirability' tex2html_wrap_inline3308 is calculated for each potential target i (i.e. food source or mate). Assuming tex2html_wrap_inline3312 where n is the number of potential targets, then the procedure for choosing the most desirable target can be specified in pseudo code (D[i] represents tex2html_wrap_inline3316 ):

for (i = 1:n)
  sort(D[i]); // D[1] has the largest value.
if (n==1 || D[1]>>D[2])
  target := 1;
else {
  K := min[N, n]; // N is a small integer. Typically, N=3.
  for (k = 1:K) {
    calculate RelPos[k]; // RelPos[k] is the relative position of (k) to fish.
                        // RelPos can have three values: left, right, front.
    p := 0.2;            // is a small weight factor.
    D[k] += p*M(RelPos[k]); // M(front) = M(FORWARD).
  for (k = 1:K)
    D[i] := max{D[k]};
  target := i;

The desirability calculation for a food source is, for simplicity, solely based on its distance to the fish. Let tex2html_wrap_inline3318 be the distance of food i to the fish, then tex2html_wrap_inline3322 where tex2html_wrap_inline3074 is the fish's perceptual range described earlier. The desirability of a mating partner will be described in Section gif.

next up previous contents
Next: From Perception to Behavior Up: The Focusser Previous: Motor Preferences
Xiaoyuan TuJanuary 1996