Evolutionary simulations of foraging agents, controlled by artificial neural networks, unexpectedly yielded oscillating node activations in the networks. The agents had to navigate a virtual environment to collect food while avoiding predation. Between generations their neural networks were subjected to mutations and crossovers in the connection strengths. The oscillations drastically enhanced the agents performance, which was due primarily to an increased switching efficacy from approach to avoidance behavior. In this paper we further analyzed networks from the last generation and found that oscillations modulated winner-take-all competition. On average the oscillations had a much higher frequency when an agent was foraging (i.e., in an appetitive state) than when it was trying to escape from a predator (i.e., in an aversive state). We suggest that also in real brains oscillation frequencies correspond with particular types of action preparations.