CARLA Multi-fidelity fusion

This repository contains several experiments for running CARLA at diffrent fidelity levels for evaluating ADS

Repository structure

Note

Need to update Repository is divided in following structure

• experiments/
  • high_low_fidelity/ (main experiments showing influence of fidelity parameters on execution time and ADS behaviour)
  • past_experiments/
    • fidelity_v1/ (past fidelity experiments on FPS and substepping)
    • fidelity_v2/ (past fidelity experiments on FPS and render quality)
    • determinism_check/ (checking if ADS is deterministic spoiler its NOT)
    • paralelism_check/ (checking how much time we can save on running two instances of CARLA and ADS at once)

Each experiment directory contains data for this round of experiments and a Python notebook that loads the data and answer questions.

Data generation

Data was obtained by evaluating InterFuser state of the art autonomous driving system (ADS).

ADS was evaluated using custom scripts to change simulator fidelity parameters.

Requirements

• Python
• Pandas
• Numpy
• Scipy
• Matplotlib

Tested with provided conda environment (environment.yml).

Contact

Please raise an issue here or send me an email:

amosikowicz1@sheffield.ac.uk