Adaptive Behaviour based on Evolving Thresholds with Feedback

In this paper, we propose a model for dynamic task selection based on an evolving threshold mechanism. The threshold evolves based on statistical reinforcement with experience the agent gets from positive and negative feedback. This gives a model where the task selection is tightly coupled with the real success of the tasks and with the perceptions of the agent in the form of stimuli. An example and experiment in the context of ant simulation shows that the agent adapts itself in a flexible way, and at the same time uses experience to achieve a simple kind of specialisation.