Publikation

ReconfROS: An approach for accelerating ROS nodes on reconfigurable SoCs

Marc Eisoldt, Marcel Flottmann, Julian Gaal, Steffen Hinderink, Juri Vana, Marco Tassemeier, Marc Rothmann, Thomas Wiemann, Mario Porrmann

In: Microprocessors and Microsystems (MICPRO) 94 Seite 104655 Elsevier 2022.

Zusammenfassung

In this paper, we present an approach to integrate reconfigurable SoCs into the well known Robot Operating System (ROS) called ReconfROS. Our method allows to implement hardware-accelerated algorithms on FPGAs and integrates them directly into the ROS ecosystem. This allows to combine the established and well tested ROS infrastructure together with low-power hardware acceleration. As a proof-of-concept for this novel integration, we ported an existing image-based path-following algorithm onto an FPGA and tested it on an unmanned ground vehicle (UGV). Evaluations utilizing the FPGA-accelerator directly synthesized from C-code show a three times performance improvement combined with a 15 times reduction of energy per frame. The approach can be easily ported to different hardware platforms based on Xilinx reconfigurable SoCs.

Deutsches Forschungszentrum für Künstliche Intelligenz
German Research Center for Artificial Intelligence