State-Space Perturbation to Assess Multi-Agent Coordination

Author(s):

Erin Zaroukian, Rolando Fernandez, and Derrik Asher

Abstract:

This report outlines state-space perturbation, a method developed to perturb the state space of a reinforcement learning agent to determine the degree to which its behavior is influenced by the behavior of other agents in its state space. Two applications are described: a predator–prey pursuit task and a turret-reconnaissance task, followed by a formal description of the method and a guide to performing state-space perturbation using internally developed code.