next up previous
Next: Quantitative Analysis of Behaviors Up: Emergent Behaviors Previous: Unforced errors

Behavior or Imagination?

Are these behaviors really occurring as we described them? Some cases, such as spiraling, are quite obvious, but others like the cutoff and the trap might be more in the imagination of the observer. After all, a Tron agent cannot decide that the time is ripe for a cutoff maneuver without knowing the position of the other player. Nor it has memory to decide: ``I am in the process of doing a spiral, so turn in the same direction again''.

On the other hand, it might be the case that the cutoff is ``definable'' for an agent in different terms than ours. As long as there is a certain environmental condition (in terms of sensory inputs) that is correlated with the right moment to do a cutoff, agents likely to turn in the correct direction in those circumstances might have an evolutionary advantage, so in the end we see agents ``doing it'' with increasing frequency.

Ethologists warn us [106, p. 105] that an animal's action is called behavior only when performed in order to improve its chances of survival or reproduction -- because it has been selected for, that is. The present section just shows a group of snapshots to illustrate the types of actions that Tron agents are performing. In the next one we go further, testing for correlations between behaviors, evolutionary time, and fitness.


next up previous
Next: Quantitative Analysis of Behaviors Up: Emergent Behaviors Previous: Unforced errors
Pablo Funes
2001-05-08