SSubhnil / RacingCARLA

Learning Model Predictive Control (LMPC) for autonomous racing in CARLA 3D environment.

Notice #2: I am back currently working on this with Carla 0.9.11 (packaged version). I had taken a long sabbatical to work on few other things. I will be updating the code, do a lot of code clean-up and will release a stable version soon. Thanks for "Starring" and "Forking".

Notice #1: After the recent Carla update 0.9.10, the code only works for the packaged version. There are errors in the source build which are being resolved. The code will work for 0.9.9 (if built from source).

CARLA-Racing

A program for a self-driving car to learn to drive faster on a racetrack. The 3D environment is CARLA, an open source autonomous driving simulator. We use vehicle sensors like, for example, RGB camera, RADAR, LIDAR, IMU, etc. for accurate state estimation and, for the time being, Learning Model Predictive Controller (LMPC) which is a data-driven learning controller.

See https://github.com/urosolia/RacingLMPC for LMPC implementation.

Requirements (CARLA Packaged):

See https://github.com/carla-simulator/carla/releases for download. You will need Python 3.7 with OpenCV, NumPy, SciPy and CvxOpt. A dedicated GPU with at least 4 GB VRAM (especially for Machine Learning). Minimum 8 GB DDR3 RAM.

1. If you are running the packaged version, and not building from source, simply download and extract the repository in the examples folder in PythonAPI.
2. Run CarlaUE4.exe. This will start the server in a game window.
3. Run commander.py in Python 3.7 and observe the game window.

Post-setup Instructions (CARLA Packaged):

Currently the simulation is coded on Town07 map. By default the Unreal Editor loads Town03. Town07 is an additional map and is not included with the base package. The additional maps can be downloaded from https://github.com/carla-simulator/carla/releases. Extract the .rar or .zip in [root-CARLA folder] and overwrite on prompt. Before starting CarlaUE4.exe, go to [root-Carla folder] -> CarlaUE4 -> Config. Open DefaultEngine.ini with Notepad. Change Town03 to Town07 for all assignments. The Unreal Editor will now load the project with Town07.

Requirements (CARLA from source):

1. CARLA built from source (Wondows and/or Linux). See https://carla.readthedocs.io/en/latest/build_windows/ for installation instructions. It is highly advisable to strictly follow the instructions for successful installation. Note that You should have both Windows 10 SDK and Windows 8.1 SDK for VS 2017.
2. Unreal Engine 4.24.3 with debugging symbols add-on. This can be done via Epic Games Launcher (https://www.epicgames.com/site/en-US/home)
3. Python 3.7 with OpenCV, NumPy, SciPy and CvxOpt.
4. Dedicated GPU with at least 4 GB VRAM (especially for Machine Learning).
5. Minimum 40 GB HDD space.
6. Minimum 8 GB DDR3 RAM

Post-setup Instructions (CARLA from source):

Currently the simulation is coded on Town07 map. By default the Unreal Editor loads Town03. Close the Unreal Editor, and exit the Epic Games Launcher. Go to [Carla-master folder] -> Unreal -> CarlaUE4 -> Config. Open DefaultEngine.ini with Notepad. Change Town03 to Town07 for all assignments. The Unreal Editor will now load the project with Town07. This can also be done from within Unreal Editor by accessing Settings -> Maps and Modes.

I need a team for the projects, so don't hesitate to get in touch via e-mail in bio.