Modelling Behaviour Cycles as a Value System for Developmental Systems
The behavior of natural systems is governed by rhythmic behavior cycles at the biological, cognitive and social levels. These cycles permit natural organisms to adapt their behavior to their environment for survival, behavioral efficiency or evolutionary advantage. This project is developing models of behavior cycles as the basis for motivated reinforcement learning in developmental systems such as virtual agents and robots. Motivated reinforcement learning is a machine learning technique that incorporates a value system with a trial-and-error learning component. This project has developed and evaluated three value systems based on behaviour cycles and three function approximation models for motivated reinforcement learning. These models have been evaluated in virtual agents and on four Lego Mindstorms NXT robots, shown in Figure 2. Results show that both the virtual agents and the robots can evolve measurable, structured behavior cycles adapted to their individual forms. These results have been accepted for publication in the Adaptive Behavior journal and presented at the inaugural International Workshop on Intrinsically Motivated, Cumulative Learning, Versatile Robots.