Nicholas L. Cassimatis, J. Gregory Trafton, Alan C. Schultz, and Magdalena D. Bugajska
We argue that many problems in robotics arise from the difficulty of integrating multiple representation and inference techniques. These include problems involved in planning and reasoning using noisy sensor information from a changing world, symbol grounding and data fusion. We describe an architecture that integrates multiple reasoning, planning, sensation and mobility techniques by composing them from strategies of managing mental simulations. Since simulations are conducted by modules that include high-level artificial intelligence representation techniques as well as robotic techniques for sensation and reactive mobility, cognition, perception and action are continually integrated. Our work designing a robot based on this architecture demonstrates that high-level cognition can make robot behavior more intelligent and flexible and improve human-robot interaction.