Simulated autonomous agents utilizing instincts and behavior learning: Orgs in Orgland

This paper introduces a simulator of interacting agents wherein an artificial agent, dubbed an Org, can learn and accumulate experience built upon initially endowed primordial instincts. Building on these instincts, the Org develops experience, the ability to detect behavioral patterns and discover physical constraints through self-reflection. Emphasis is placed on the concurrent use of long-term planning and reactive behavior. The interaction of the Org with its environment takes place in a three dimensional world where pursuit-evasion games are conducted with other Orgs. The Org is an autonomous agent capable of learning its environmental constraints and self-organizing knowledge and experience. The long-term goal of this project is to create a generic artificial agent, used as an unsupervised problem solver on a diverse set of environments.