ROS 2 Latency Analysis

This repository contains the source code used in the paper "Latency Measurement for Autonomous Driving Software Using Data Flow Extraction". It constructs a data flow graph (DFG) for a ROS2 autonomous driving software using a static code analysis. Therefore, intra-node data dependencies of C++ based ROS2 system can be found without manual annotation.

The DFG is annotated with tracing data from ros2_tracing where we can extract different latency variables:

We define the end-to-end latency as the time between receiving the input data from the first node and sending the output values in the last node of the considered compute graph.
The communication latency is the time needed to send messages between the ROS2 nodes. It is the time between publishing and receiving via a subscription.
The compute latency defines the time between the subscription of a topic, and the point where it publishes the processed data to the following node.
The idle latency defines the time taken for intra-node communication i.e., between subscription and timer callback.

The framework is applied in "How Fast is My Software? Latency Evaluation for a ROS2 Autonomous Driving Software", where we analyzed the Autoware software stack. The entire simulation setup can be found in this repository.

Setup

This repository contains tools to

execute ROS 2 programs repeatably with tracing and resource usage tracking
extract ROS 2 node-internal data dependencies from C++ source code
analyze end-to-end latencies and their breakdowns, given data from 1. and 2.

The tools are found in the branches of this repository.

Runner Framework

For data acquisition / simulation runs, up to three hosts can be utilized. For all those hosts, the setup is as follows:

    git clone -b runner-framework https://github.com/TUM-AVS/ros2_latency_analysis.git runner_framework
    cd runner_framework
    ./setup.bash

Code Analysis

C++ dependency analysis has to be done on the same code base that is run in the data acquisition step but the machine can differ. The setup is as follows:

    git clone -b code-analysis https://github.com/TUM-AVS/ros2_latency_analysis.git code_analysis
    cd code_analysis

Follow the instructions in the README.md file in that directory.

Dataflow Analysis

Dataflow analysis has to be done with the same ROS 2 version as the above steps but the machine can differ. The setup is as follows:

    git clone -b dataflow-analysis https://github.com/TUM-AVS/ros2_latency_analysis.git dataflow_analysis
    cd dataflow_analysis

Follow the instructions in the README.md file in that directory.

Paper Reference

If you find the information in this repository useful we would be happy if you cite it based on the following definition:

T. Betz, M. Schmeller, H. Teper, and J. Betz, “How Fast is My Software? Latency Evaluation for a ROS 2 Autonomous Driving Software,” 2023 IEEE Intelligent Vehicles Symposium (IV). IEEE, Jun. 04, 2023. doi: 10.1109/iv55152.2023.10186585.

@inproceedings{Betz2023,
  doi = {10.1109/iv55152.2023.10186585},
  url = {https://doi.org/10.1109/iv55152.2023.10186585},
  year = {2023},
  month = jun,
  publisher = {{IEEE}},
  author = {Tobias Betz and Maximilian Schmeller and Harun Teper and Johannes Betz},
  title = {How Fast is My Software? Latency Evaluation for a {ROS} 2 Autonomous Driving Software},
  booktitle = {2023 {IEEE} Intelligent Vehicles Symposium ({IV})}
}

T. Betz, M. Schmeller, A. Korb, and J. Betz, “Latency Measurement for Autonomous Driving Software Using Data Flow Extraction,” 2023 IEEE Intelligent Vehicles Symposium (IV). IEEE, Jun. 04, 2023. doi: 10.1109/iv55152.2023.10186686.

@inproceedings{Betz2023,
  doi = {10.1109/iv55152.2023.10186686},
  url = {https://doi.org/10.1109/iv55152.2023.10186686},
  year = {2023},
  month = jun,
  publisher = {{IEEE}},
  author = {Tobias Betz and Maximilian Schmeller and Andreas Korb and Johannes Betz},
  title = {Latency Measurement for Autonomous Driving Software Using Data Flow Extraction},
  booktitle = {2023 {IEEE} Intelligent Vehicles Symposium ({IV})}
}