The goal of this venture is to develop models capable of completing a variety of autonomous tasks within the Carla simulator using reinforcement learning methods. The repository is divided into two distinct projects which are still being developed.
Autonomous ride from point A to B. Trained models (see A_to_B/final_models) are able to correctly perform fundamental road manoeuvres such as driving straight, turning left and right. Using this knowledge, the models can learn how to beat different scenarios in a reasonable amount of time, depending on the scenario's complexity.
Examples of some fundamental road manoeuvres:
Example of a more challenging scenario:
Autonomous chase of a fleeing vehicle. This project followed a similar approach to learning. However, autonomous chasing proved to be more challenging than getting from point A to point B, and work on training a satisfactory model continues.
Examples of some fundamental chase manoeuvres:
- Provide the paths to the Carla executable and egg files in the file settings.py,
- Run a2c_rgb.py file.
- Python 3.7.x
- Pipenv
- Git
- Carla 0.9.10
The pipfile contains a list of all essential project packages. For information on how to install these, see the setup section. If you wish to use a GPU with Pytorch, you'll need a device that supports CUDA.
- Install git, python and pipenv
- Clone this repository and navigate to its root directory
git clone /~https://github.com/Michal-Kolomanski/Autonomous-driving-in-Carla- Install all required project packages by executing
pipenv install --dev- To open project virtual environment shell, type:
pipenv shell