← Research log
SafeGraph-CAVexactnegative result

SUMO/TraCI setup and the netgenerate failure

SUMO was installed in the Colab/Linux environment.

Progress Update

SUMO was installed in the Colab/Linux environment. The sumo --version command worked and reported version 1.12.0. SUMO_HOME was configured as /usr/share/sumo, the tools path was added, and TraCI imported successfully from /usr/share/sumo/tools/traci/__init__.py.

Experiment Note

The netgenerate --version command crashed with exit code 139. Rather than spending the project window debugging the distribution-specific binary failure, the road network was written manually in SUMO XML. A small straight-road scenario was created, including the A0B0 edge, routes, vehicles, and a SUMO configuration file. This bypassed netgenerate without changing the experimental question.

The planned runner was scripts/sumo/run_sumo_platoon.py. The initial SUMO matrix covered attacks none, fdi_position, fdi_velocity, delay, and packet_loss, with outputs written to CSV for later aggregation.

Technical Essay

SUMO was added as an external microscopic traffic simulator, not as the primary place to develop every part of the method. The lightweight simulator remained valuable because it made attack generation, paired evaluation, and large experiment sweeps cheap and controlled. SUMO added a second implementation layer that could expose integration errors and confirm whether the prevention pattern survived a more realistic traffic execution environment.

The decision to manually define a minimal network was intentionally pragmatic. The paper did not need a complicated road topology; it needed an independent simulator and TraCI-controlled vehicles. A straight two-lane road was sufficient to test the longitudinal cyber-control mechanism.

IMPLEMENTATION AND EXPERIMENTAL DEVELOPMENT

This reconstruction preserves the source range and editorial confidence label. An approximate date or empirical result is never strengthened during editing.

Related case study →