Some Models for Autonomous Agents' Action Selection in Complex Partially Observable Environments

We study the resource-bounded autonomous agents acting in complex multi-agent and multi-task environments. In particular, we investigate various models of the agent action selection in such environments. Designing such autonomous decision making agents for non-episodic and partially observable dynamic environments is particularly challenging, due to a high-level demand such environments pose in terms of agent’s necessary capabilities. We make an early attempt in modeling and designing agents for largescale multi-agent systems and complex environments, where individual agent’s behaviors are sufficiently simple to be scalable and practical, and where agents’ coordination and self-organization capabilities can still make agents (both individually and as ensembles) effective with respect to accomplishing their goals. The emphasis in this paper is on models of an agent’s local knowledge based individual behaviors. We propose several simple mathematical models for an agent’s local knowledge-based action selection. We illustrate the general ideas about bounded-resource autonomous agents acting in multi-agent, multi-task dynamic environments, and the proposed generic models of agents’ autonomous local-knowledge based action selection in such environments, with a concrete application example: modeling and simulation of a collection of autonomous unmanned aerial vehicles (UAVs) on a multi-task mission.