Real-time machine learning in embedded software and hardware platforms

Abstract

This paper describes on-going research work into real-time machine learning using embedded software and reconfigurable hardware. The main focus of the work is to develop real-time incremental learning methods particularly targeted at demonstration in mobile robot environments. Three main areas are described. The first represents reactive robot navigation knowledge using a novel frequency table technique whose memory requirement is known a priori. The second area investigates a Genetic Algorithm (GA) method that combines planning and reactive approaches to allow navigation to proceed even in the face of time constraints. In the third area we are developing novel hardware-based machine learning systems suitable for implementation in reconfigurable platforms.