In this paper, we propose the combined use of reinforcement learning and game theory to provide autonomous, distributed control of a Tactical Sensing Grid (TSG) consisting of multiple sensor platforms. This paper builds upon our previously reported successful, generalizable control of a single multi-modality sensor platform using online reinforcement learning. We use game theory to solve an assignment problem allocating sensor platforms to known targets in a scenario. We formulate this assignment problem as a potential game, a type of game for which multiple algorithms are known to converge to Nash equilibria. Our formulation guarantees that, under mild assumptions, all Nash equilibria result in all known targets being allocated to a given sensor. We leverage this potential game formulation to propose a method for optimizing situational awareness in a TSG that is robust to dynamic network topologies, where online reinforcement learning is used to control each individual sensor platform and game theory is used locally (in the geospatial sense) to allocate nearby sensor platforms to known targets in their vicinity. We present simulation results to illustrate how sensor node negotiations lead to near-optimal assignments of sensors to targets as scenarios evolve over time.
Multiagent Reinforcement Learning and Game-Theoretic Optimization for Autonomous Sensor Control
02.03.2024
1521024 byte
Aufsatz (Konferenz)
Elektronische Ressource
Englisch
Game-theoretic payoff allocation in multiagent machine learning systems
BASE | 2022
|Feudal multiagent reinforcement learning for interdomain collaborative routing optimization
BASE | 2022
|Taylor & Francis Verlag | 2024
|