carla-rl

• Reinforcement Learning implementations for Carla simulator https://github.com/carla-simulator/carla.

• So far, it is implemented the dueling deep-Q learning with prioritized experience replay

• Tested on carla version 0.9.5

How to run

• Clone the project inside PythonAPI examples folder: carla_folder/PythonAPI/examples
• Edit rl_config.py with the necessary hyperparameters
• Run rl_agent.py. The possible arguments are listed below
• After training, the resulting model can be tested by running rl_agent.py --test

rl_agent arguments

The arguments are adapted from manual_control.py, with minor changes

• '--test': test a trained model
• '-v', '--verbose': print debug information
• '--host' (default='127.0.0.1'): IP of the host server
• '-p', '--port' (default=2000): TCP port to listen to
• '-a', '--autopilot': enable autopilot
• '--res' (default='800x600'): window resolution
• '--filter' (default='vehicle.audi.tt'): 'actor filter
• '--rolename' (default='hero'): actor role name