Advancing ROS 2 for Reliable Real-Time Systems: From Time-Critical Synchronization to Adaptive Resource Management

Abstract

As the successor to the Robot Operating System (ROS), ROS 2 is a pivotal framework for developing complex, real-time robotic systems. Although widely adopted, ROS 2 faces critical challenges in ensuring deterministic timing guarantees and predictable resource management, undermining reliability across applications such as autonomous vehicles and collaborative robotics. This thesis addresses three foundational gaps in ROS 2 through formal analysis and adaptive resource management architecture.

First, we target on the message synchronization, a core mechanism for aligning multi-sensor data in distributed ROS 2 based systems. By formally modeling the message synchronization policy in ROS 2, we derive a theoretical bound for the worst-case time disparity, the maximal timestamp difference among synchronized sensor inputs. Experimental validation confirms the tightness of the bound and demonstrates the advantages of ROS 2 over the alternative in minimizing temporal discrepancies.
Second, we extend the analysis to quantify two critical latency metrics: passing latency (time between message arrival and departure) and reaction latency (responsiveness to new messages). This provides predictable latency guarantees for designing real-time systems, validated through empirical testing under diverse workloads. Together with the first contribution, these insights enable developers to systematically optimize the temporal consistency in sensor coordination.

Third, we address uncoordinated GPU resource usage in ROS 2, a widespread issue in compute-intensive ROS 2 applications. We propose ROSGM, a real-time GPU management framework for ROS 2, which supports plug-in policies for dynamic task loading/unloading. ROSGM provides adaptive task execution management to eliminate rigid resource allocation, enhancing computational efficiency and maintaining flexibility for evolving operational requirements.

By bridging theoretical analysis with practical frameworks, this thesis advances the predictability, efficiency, and adaptability of ROS 2 systems, offering insights and tools applicable to timing-critical robotic applications, such as industrial automation, assistive robotics, and autonomous vehicles.

Date of Award	16 Jul 2025
Original language	English
Awarding Institution	City University of Hong Kong
Supervisor	Nan GUAN (Supervisor)