TILDE: Topic-Tracking Infrastructure for Dynamic Message Latency and Deadline Evaluator for ROS 2 Application

Abstract

Autonomous-driving systems require real-time processing for safety. Detecting the deadline miss (disabling to execute the process until the designated time) of autonomous-driving software is essential for executing real-time processing. We proposed the application of a dynamic message tracking system (Topic-tracking Infrastructure for Dynamic Message Latency and Deadline Evaluator) to software based on Robot Operating System (ROS) 2, e.g., Autoware (autonomous-driving software). TILDE transmits MessageTrackingTag with ROS 2 messages to trace and dynamically specify the data flow of the messages. In particular, MessageTrackingTag contains input and output times at each node with the data flow. The information of the MessageTrackingTag is acquired by a deadline detector to identify the time from the input to output (hereafter latency) of the data flow. The latency is compared to the deadline to detect the occurrence of a deadline miss. In principle, the deadline detector is a package of TILDE for notifying deadline misses. In addition, this paper conducted two evaluations of TILDE. First, we measured the overhead caused by embedding TILDE on ROS 2 nodes through the transmission of a String-type message from one node to another. Second, we measured the deadline miss detection rate using CARET (a latency analysis tool for autonomous-driving systems) to validate the accuracy of the deadline detector in determining deadline misses.