7.3.4 Goal-directed movements

The goal for the robot is to move in a way that the sensory value in a single given sector reaches a given value. Figure 7.12 illustrates the result of two typical movements. In example A, the goal was to make the sector in the back right (number 4 in figure 7.3, right) attain a low value (for example, 50 pixels). The robot moved backward in a rightward curve. In example B, the front sector should attain a low value. Thus, the robot moved from the middle of the circle straight toward the obstacles.

Figure 7.12: Typical goal-directed movements. The goal was to be close to an obstacle (A) in the back right sector (number 4 in figure 7.3) or (B) in the front sector (number 0 in figure 7.3).

The two optimization methods gave similar results (table 7.3). In almost all trials, both optimization methods found a solution (96% to 99%). In 15% to 18% of the trials, the final sector value matched exactly the desired value, and in almost half of the trials, the final value was within one pixel of the desired value. In more than 80% of the trials, the final sector value was closer to the desired than to the initial value, and in more than 90% of trials, the final sector value was changed in the right direction (increasing or decreasing).

Table 7.3: Result of 100 goal-directed movements for simulated annealing and Powell's method.

	sim. anneal.	Powell
found solution	99%	96%
exact hits	18%	15%
close within one pixel	41%	46%
closer to goal	85%	83%
right direction	91%	91%
mean square error/link	4.4	2.6
mean number of links	2.8	2.5

The almost equal performance of the Simulated Annealing and the Powell's method did not depend on choosing just one goal sector. No noticeable difference was also observed when two and three defined goal sectors were defined (data not shown).