Overcoming allostatic challenges through predictive robot regulatory behavior

Abstract

Internal processes such as homeostasis and allostasis operate to keep the internal environment within desired conditions to sustain fitness by satisfying rising needs such as thirst or hunger. However, when two or more needs are to be satisfied, the organism faces a conflict and based on diverse factors, from interoceptive sensations to external stimuli from the environment, one of the needs is prioritized and satiated over another. Allostasis, as a predictive mechanism, is at the core of effective regulation and conflict resolution. In this work, we simulate competing emerging needs such as thirst and internal temperature by adding a feedforward module (Allostasis), responsible for the predictive behavior of a simulated agent over an already existing model of reactive homeostasis, in which the agent is placed within an environment of constantly changing temperatures. Incorporating the anticipatory layer happens at two conditions, single and multiple drive prediction, and it is hypothesized that the agent under the predictive conditions will have less homeostatic error over time compared to the reactive one. The results show a significant reduction of homeostatic error on both conditions upon the addition of the feedforward controller, supporting and contributing to the literature on allostatic anticipation and effective regulatory control. Moreover, methodological recommendations for further research are given based on the limitations found in the development of this study.