Mini-Robot's Performance Optimization via Online Reconfiguration and HW/SW Task Scheduling

Abstract

Nowadays a big effort is spent on research in heterogeneous computing and in particular in accelerator-based solutions. The field of embedded systems and in particular researches regarding autonomous robots have a great interest in this subject since the use of FPGAs on these devices as coprocessors is a common practice due to the tasks they have to accomplish. Autonomous robots are used to perform environment analysis, for example in critical situations, when human intervention could be dangerous or even impossible. In this paper in order to exploit their potentialities, given their heterogeneous nature, techniques from hardware and software codesign have been applied to this field. A particular focus during the design has been given to the problematic of runtime reconfiguration in order to efficiently use them in real scenarios. This work presents a methodology which permits, given multiple implementations of the same task, to choose at runtime which one to use on the basis of off-line profiling informations in order to achieve the best execution time. This research has been carried out by implementing such a technique in the case of a robot equipped with an ARM processor running GNU/Linux, and a Xilinx FPGA as a hardware accelerator.