Towards an Adaptive Levy Walk using Artificial Endocrine Systems

Behavioural adaptation is often observed in foraging animals via coupling periods of localized search and long straightforward motions, a well-known strategy named Lévy Walk. In this paper we propose an adaptive Lévy Walk model to control an autonomous agent. The model is comprised of a Lévy-based controller modulated by an artificial endocrine system optimised through evolutionary techniques. This new approach enables the agent to control the transition between localized search and long relocation when exposed to external stimulus. The model is tested in exploration tasks where environments have resources clustered into patches. Further tests incorporated environments with different patch characteristics, such as patch size or resource distribution within patches. Our model has shown to outperform the benchmark approach in terms of search efficiency, highlighting the benefits of combining a Lévy Walk based controller with a biologically inspired strategy for adaptation.