Overview of the dynamic predictive architecture for robotic assistants
Charles E. Martin, R. James Firby · Proceedings of SPIE, the International Society for Optical Engineering/Proceedings of SPIE · 1992
A robotic assistant for extra-vehicular activity in space must deal with a complex, constantly changing environment. Classical planning architectures are ineffective for such tasks, but modern reactive systems are a plausible alternative. In addition, the robot must be capable of effective real-time response to the needs of human astronauts. Writing and debugging robot programs will not be possible under actual work conditions; astronaut-assistant dialogues must, therefore, take place in natural language. Unfortunately, the generally nonrepresentationalist nature of reactive systems make generative natural language interfaces impossible. In this paper, we present an overview of the Dynamic Predictive Memory architecture for robotic assistants. This architecture is an extension of the Direct Memory Access Parsing (DMAP) model of language understanding, in which the data structures and algorithms associated with the reactive action package (RAP) execution system are represented in the uniform memory format of the system. This allows natural language reference to take place coincident with the reactive execution of plans. The result is a reactive system which human users can interact with in natural language.