Program Optimization for Real-Time Performance

Abstract

In this chapter, we present methods that allow a robot to use high-level knowledge for reasoning about its actions in real time. The knowledge is represented by a logic program that contains learned relational rules. In order to operationalize this knowledge we develop program optimization methods that speed up its evaluation. These are a program transformation method, a compilation method, and an efficient forward inference method. The latter is part of a performance system capable of inferring high-level concepts from the sensory input of a mobile robot on-line and in real time.