SimLKAS

Code for paper:
SimLKAS: a simulation-based framework for verification and validation of lane keeping assistance systems.
This repository accompanies the publication:
https://doi.org/10.1080/15472450.2025.2559224

---

Description

SimLKAS provides a modular simulation framework to evaluate LKAS performance under various conditions, including adverse weather and complex road geometries. It integrates perception, control, and evaluation components for systematic testing.

---

Project layout

• run_sim.py: single entry point to run demos or experiments from JSON configs.
• configs/: example configs (demo.json, experiment.json) and experiment.schema.md documenting fields.
• modules/: code for simulator (simulator/runner.py), controller, perception (legacy + LaneNet), HUD, and shared utils.
• lanenet/: LaneNet model code and weights (testing/benchmark only).
• results/: default output directory for trajectory CSVs when logging is enabled.

Quick start

1. Prereqs: a Python env with CARLA Python API, pygame, OpenCV, TensorFlow (only for lanenet), scikit-learn, numpy.
2. Start a CARLA server (e.g., CarlaUE4.sh -carla-port=2000), default host/port localhost:2000.
3. Run with a config:
   • Demo (single run, HUD on, no logging):

     python run_sim.py -c configs/demo.json

   • Experiment (multi-run, logging on):

     python run_sim.py -c configs/experiment.json

4. To customize, edit a JSON per configs/experiment.schema.md. Key fields:
   • experiment.mode: single or multi
   • environment.weather: preset name from configs/weather_presets.json
   • lkas.detector: legacy (no TF) or lanenet (TF required)
   • display.show_hud: true/false
   • logging.enabled: true/false

Output naming

When logging is enabled, trajectories are written to results/ as:

<detector>_<map>_<weather>_<target_speed>_<street_light>_<vehicle_light>[_seedN]_trajectory.csv

_seedN is included only when a seed is supplied.

---

LaneNet Model Notice

This repository includes the LaneNet model code, copied from
MaybeShewill-CV/lanenet-lane-detection.

• LaneNet is only used for testing purposes in our experiments.
• It is not the backbone of our lane detection algorithm within SimLKAS.
• Full credit for LaneNet goes to the original authors. Please see their repository for details and licensing.

---

Citation

If you use this framework in your research, please cite:

Yang, T., Ding, Y., Li, K., Pan, Y., Qin, R., Yin, Z., & Hu, X. (2025).
SimLKAS: a simulation-based framework for verification and validation of lane keeping assistance systems.
Journal of Intelligent Transportation Systems, 1-25.
https://doi.org/10.1080/15472450.2025.2559224