# ROS 2 Trace Analysis for Semantic Scheduling (Thesis)

This repo provides the data analysis framework used for my thesis  
**"Semantic Scheduling for Multi-Modal Sensor Data on the Example of ROS 2"** (TU Berlin & DLR, 2025).

It is based on [TUM-AVS/ros2_latency_analysis](https://github.com/TUM-AVS/ros2_latency_analysis) (`dataflow-analysis` branch) by Betz et al.  
**Please cite their original work when using this code (see below).**

## Key Features

- Automated extraction of end-to-end and component latencies from ROS 2 traces
- Extended batch processing, including meta-anaylsis
- Scripts and fixes tailored for semantic scheduling experiments

## Main Additions & Changes

- **New scripts:**  
  `add_csv_header.sh`, `batch_analysis_analysis.py`, `csv2table.py`, `csvfix.sh`,  
  `run_batch_analysis_analysis.sh`, `run_batch_analyze.sh`
- **Main modifications:**  
  `batch_analyze.py`, `trace-analysis.ipynb`
- **Submodules:**  
  Includes `ros2_tracing` and `tracetools_analysis`

All changes are documented in the git history.

## Setup

- Requires **Python 3.10+**, ROS 2, JupyterLab, and [ros2_tracing](https://github.com/ros2/ros2_tracing)
- Install dependencies:  
  `pip3.10 install -r requirements.txt`
- Run interactive analysis in Jupyter (`trace-analysis.ipynb`) or batch via provided scripts.

See the [upstream README](https://github.com/TUM-AVS/ros2_latency_analysis/tree/dataflow-analysis) for more details on prerequisites and methodology.

## Usage

First, make sure all your traces are in sub-directories conforming to the common experiment type, e.g.:
```
traces/
  ros_multi/
    ros_multi_1
    ros_multi_2
    ...
    ros_multi_n
  ros_single/
    ros_single_1
    ros_single_2
    ...
    ros_single_m
```
Then use `run_batch_analyze.sh` to run the analysis (trace-analysis.ipynb) for the traces you gathered. This will create a summarizing `results.csv` in the respecitve directory (here thus `traces/ros_multi/` and 'traces/ros_multi/').  
The csv will not have headers, to add them, use `add_csv_header.sh`.  
Now you can run `run_batch_analysis_analysis.sh` to generate the end-to-end analysis box-plots from that results.csv (will also generate a `results_summary.csv`). The latter is ideal to be turned into LaTeX tables, using `csv2table.py`.

## Output

- Latency CSVs (E2E and per-chain), plots (PDF/PNG), and LaTeX-ready tables

## Citation

If you use this code, **please cite**:

Betz, T. et al., “Latency Measurement for Autonomous Driving Software Using Data Flow Extraction,”  
*IEEE Intelligent Vehicles Symposium (IV), 2023.*  
[DOI: 10.1109/iv55152.2023.10186686](https://doi.org/10.1109/iv55152.2023.10186686)

```bibtex
@inproceedings{Betz2023,
  doi = {10.1109/iv55152.2023.10186686},
  url = {https://doi.org/10.1109/iv55152.2023.10186686},
  year = {2023},
  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})}
}
```

## License

See upstream for original license; thesis-specific additions are for academic/non-commercial use.