ROSOMP: Coordinated Scheduling of OpenMP Parallel Regions in ROS 2 Executor

Abstract

The Robot Operating System (ROS 2) is widely adopted for modular robotic applications. OpenMP provides an approach for parallelizing CPU-bound workloads, and is commonly used in ROS-based applications for perception and control. However, integrating OpenMP into the ROS 2 executor introduces semantic mismatches: OpenMP threads are managed independently by the OS scheduler, breaking ROS-level priority semantics and leading to priority inversion, thread oversubscription, and timing unpredictability. This article presents ROSOMP, a unified framework that enables priority-driven scheduling and bounded execution of OpenMP parallel regions within the ROS 2 runtime. By converting OpenMP constructs into schedulable units and executing them within the existing ROS 2 executor thread pool, ROSOMP enforces global priority consistency while preserving the default programming model of ROS 2 and OpenMP. Furthermore, we develop a response time analysis to formally establish the predictability of ROSOMP. Experimental results demonstrate that ROSOMP significantly improves timing determinism and reduces system overhead compared to conventional OpenMP integration in ROS 2.