A Habit System for an Interactive Robot

摘要

Human behavior is a mix of automatic perceptually-triggered habits and intentional goal-directed actions. We have implemented a habit system that automatically executes actions based on internal and external context. The result of the habit system is a homeostatic mechanism with implicit goals that balances the physical needs of a physical robot with other factors such as interaction with human partners. By acting automatically on salient context elements, the robot intersperses the use of its conversational interface with self-preservation and curiosity drives. Ultimately, having both a goal and a habit system governing robot behavior will allow a very rich platform for embodied, grounded semantics. We present details of the model and implementation, and discuss insights into the embodied nature of semantics that arises from this work.