Incremental Signal to Symbol Processing

Abstract

Acquiring symbolic descriptors from raw sensor data is an important task in two respects: first, it is a prerequisite for logic-based learning. The resulting descriptors form the language for the examples of further learning. Second, it is a prerequisite for higher-level planning. The planning component uses the resulting descriptors instead of the numerical sensor measurements. Hence, the signal to symbol processing bridges the gap between numerical sensor data and symbolic approaches to both learning and planning. In this chapter, we present an approach to transforming a stream of numerical sensor measurements into a sequence of symbolic descriptions of the sensor measurements. This approach is strongly incremental: no measurements need to be stored. The compression into symbolic descriptions considers only the symbolic description gained so far and the current measurement. The incrementality allows on-line processing of sensor data in real time. Another characteristic of the approach is that only sensor and movement measurements are given. No map or position information is required. The approach has been applied to the navigation of the mobile robot PRIAMOS.