Generalized Ising model for dynamic adaptation in autonomous systems

Abstract

The paper presents a concept of Statistical Mechanics for observation-based adaptation in autonomous systems, which is typically exhibited by simple biological systems. Time-critical operations of autonomous systems (e.g., unmanned undersea vehicles (UUVs)), require in situ adaptation in the original plan of action and rapid response to evolving contextual changes and situation awareness for enhanced autonomy. In this regard, a concept of dynamic plan adaptation (DPA) is formulated in the setting of a generalized Ising model (e.g., the Potts model) over a discretized configuration space, where the targets (e.g., undersea mines) are distributed. An exogenous time-dependent potential field is defined that controls the movements of the autonomous system in the configuration space, while the decision-theoretic tool for dynamic plan adaptation is built upon local neighborhood interactions. The efficacy of the DPA algorithm has been evaluated by simulation experiments that demonstrate early detection of localized neighborhood targets as compared to a conventional search method involving back and forth motions.