Extended ramp goal module: Low-cost behaviour arbitration for real-time controllers based on biological models of dopamine cells

The current industrial focus in virtual agents and digital games is on complex systems that more accurately simulate the real world, including cognitive characters. This trend introduces a multitude of control parameters generally accompanied by high computational costs. The resulting complexity limits the applicability of AI in these domains. One solution to this problem is to focus on light-weight flexible AI architectures which can be simultaneously generated, controlled and run in parallel. The resulting systems should then be able to control individual game characters, scaling up to large numbers of characters, forming even complex social systems. Here we contribute one element of such a system: a light-weight systems-engineering approach for enriching behaviour arbitration in action selection. Our mechanism - ERGo - improves high-level goal arbitration in existing light-weight action selection mechanisms. ERGo provides easy and reliable non-deterministic control of goal switching, activation and inhibition, allowing natural behaviour maintenance. This mechanism can aid agent design in cases where static, linear, predefined priorities are undesirable. The model underlying our approach is biomimetic, based on neuro-cognitive research on the dopaminic cells responsible for controlling goal switching and maintenance in the mammalian brain. We demonstrate and evaluate our mechanism in a real-time, game-like simulation environment, using a previously-published system as a baseline for comparison. We demonstrate that ERGo is effective, and betters the previous approach.