Skip to main content

Project Structure

This page summarizes the repository layout and points you at the most important modules and classes to inspect when working on AssettoCorsaRL.

Repository tree​

assetto-corsa-rl/
┣ configs/
┃ ┣ ac/
┃ ┗ car-racing/
┃   ┣ env_config.yaml
┃   ┣ model_config.yaml
┃   ┗ train_config.yaml
┣ scripts/
┃ ┣ ac/
┃ ┃ ┣ record_racing_line.py
┃ ┃ ┣ vis_env_observation.py
┃ ┃ ┗ vis_racing_line.py
┃ ┗ car-racing/
┃   ┣ load.py
┃   ┣ load_vae.py
┃   ┣ train_car_racing.py
┃   ┗ train_vae.py
┣ src/
┃ ┗ assetto_corsa_rl/
┃   ┣ model/
┃   ┃ ┣ noisy.py
┃   ┃ ┣ sac.py
┃   ┃ ┣ vae.py
┃   ┃ ┗ __init__.py
┃   ┣ train/
┃   ┃ ┣ train_core.py
┃   ┃ ┣ train_utils.py
┃   ┃ ┗ __init__.py
┃   ┣ ac_env.py
┃   ┣ ac_send_actions.py
┃   ┣ ac_telemetry_helper.py
┃   ┣ env_helper.py
┃   ┣ load.py
┃   ┗ __init__.py
┣ tests/
┃ ┣ test_ac_env.py
┃ ┣ test_ac_reward.py
┃ ┣ test_env.py
┃ ┣ test_sac.py
┃ ┣ test_xbox_buttons.py
┃ ┗ test_xbox_controller.py
┣ pyproject.toml
┣ README.md
┗ requirements.txt
┗ setup.cfg

Key objects 🔍​

Environment​

  • Class: assetto_corsa_rl.ac_env.AssettoCorsa
  • What to inspect: environment lifecycle, observation & action spaces, and how racing lines are loaded via AssettoCorsa._load_racing_line.
  • File: assetto-corsa-rl/src/assetto_corsa_rl/ac_env.py

Trainer / training core​

  • Class: assetto_corsa_rl.train.train_core.Trainer
  • What to inspect: env stepping, buffering, and learning updates (see _do_update).
  • File: assetto-corsa-rl/src/assetto_corsa_rl/train/train_core.py

Recorder / Visualizer​

  • RacingLineRecorder: records laps and writes racing_lines.json.
  • RacingLineVisualizer: 3D visualization and load_racing_data helper.
  • Files: assetto-corsa-rl/scripts/ac/record_racing_line.py, assetto-corsa-rl/scripts/ac/vis_racing_line.py

AC plugin & telemetry​

  • Plugin hooks: AC_RL.third_party.stdlib.AC_RL.acMain, acUpdate, handle_input_data — plugin that sends telemetry JSON via UDP and accepts commands.
  • Telemetry helper: AC_RL.telemetry.hexdump — simple UDP listener/parser useful for debugging.
  • Files: AC_RL/third_party/stdlib/AC_RL.py, AC_RL/telemetry.py

Data / checkpoints​

  • Racing line dataset: racing_lines.json — top-level JSON, contains num_laps and laps[].positions[] (timestamp, x, y, z).
  • Experiment logs: wandb/ run directories.
  • Checkpoints: models/ (e.g., sac_checkpoint_100000.pt).

Runner / policy loader​

  • Entry: assetto_corsa_rl.load.main — loads an actor, calls env.step, and handles actor outputs (action shaping/sampling are implemented here).
  • File: assetto-corsa-rl/src/assetto_corsa_rl/load.py

Tests​

  • Core env tests: assetto-corsa-rl/tests/test_ac_env.py
  • Other unit tests: assetto-corsa-rl/tests/ (look for module/unit tests here).

Quick tips​

  • To run tests: pytest assetto-corsa-rl/tests.
  • To inspect the trainer loop quickly: open train_core.py and search for _do_update.
  • To visualize recorded lines: run scripts/ac/vis_racing_line.py (requires display dependencies).