Information Homeostasis as a Nash Equilibrium: A Potential Game for Decentralized Multi-Agent Cave Exploration

Avi Soval, Keenan Albee

Keywords: Multi-robot systems, decentralized control, potential games, exploration, communication networks, autonomous robots, connectivity constraints

Abstract: The coordination of multi-robot teams in communication-constrained subterranean environments is difficult without persistent centralized control or a shared global map. We suggest a decentralized approach inspired by models of biological collective intelligence, and introduce a scalar stress signal - which captures local disagreement in belief states between agents - that can be used to guide the team to an information-rich state. We formulate a multi-agent cave exploration mission as a potential game with time-varying frontier resources and agent utility functions that target information gain, congestion reduction, and belief consensus through this stress coupling. By the convergence properties of standard potential games, we hypothesize that evolution of the system purely via best response dynamics may approach a Nash equilibria that resembles an 'information homeostasis' - where disagreement and uncertainty are uniformly low while coverage is still satisfied. We have currently implemented a base-uncoupled game without the belief consensus that admits a canonical potential game form, and will examine if the perturbed, coupled, information-focused game still possesses any NE convergence guarantees.

Submission Number: 42