Emotion-based parameter modulation for a hierarchical mobile robot planning and control architecture

Abstract

Autonomous robots are presently unable to match the adaptive capabilities of even the simplest of animals. Affective processes such as emotions are highly effective facilitators of adaptive behavior in humans and animals. Thus, it can be argued that emotions can bestow robots with similar adaptive advantages. In particular, artificial emotions can improve a robot's performance by modulating actions, prioritizing goals and providing reinforcements for learning. A hierarchical robot architecture that incorporates reactive and deliberative emotions has been developed to test this hypothesis. Reactive emotions arise from predictions of emotion-eliciting events from short-term sensor data and internal representations. Deliberative emotions are modeled as learned associations between environmental states and previous emotion-eliciting events. Emotions interact with multiple architectural levels by modulating parameters controlling the robot's degree of bias towards various competing drives, such as goal-seeking, safety and exploration. The model has been implemented on a simulated mobile robot for a navigation task.