CarIntersect environment

This package contains the CarIntersect environment.

CarIntersect simulate four way crossroad with some physics and bot cars. As a feature our environment has photo-realistic image and modifiability with a set of configs.

We create script to control car by hands. Just run run.sh in linux terminal. If you want more complicated settings run python3.6 visual_player.py with proper flags (them can be easily read from visual_player.py).

For requirements see requirements.txt from main part of project.

Folder structure:

• CarRacing_env
  • car.py - car class
  • contact_listner.py - class with contact listener for pybox2d
  • cvat_loader.py - class with utility functionality for loading xml cvat markup
  • rewards.py - class with reward policy, coefficient are setted via settings file
  • utils.py - class with data loader, this class contain images and provide instruments for work with external data sources
  • environment.py - main class of this package, provide environment with gym-like interfaces
• env_data - folder with background images, its cvat markup, car images ect
• configs - folder with basic configs
• common_envs_utils - set of files with some code relevant to env.
  • batch_evaluater.py - contain class, that create small set of environments for test evaluation
  • env_evaluater.py - set of function for single env evaluation
  • env_makers.py - some functions for creating env with specific config file
  • *_wrappers.py - files with env wrappers
  • visualizer.py - some dirty functions for work with animations
• visual_player.py - runnable .py file, that can be used for running env and controlling car from keyboard

Config structure: Config for env is .yaml files and consist of three parts:

• reward part - coefficients for reward policy described in rewarder.py. link to example
• state part - like the behavior of env: list of vector state to use for an agent, use of not picture as a state, number of bots, agents and bots tracks. link to example
• path part - paths to car images, the path to the back image, and cvat file. link to example

The final config is just the concatenation of these parts. (dict.update in python) link to example

CVAT file is used for creating a track line, track polygons, and some other.

This environment was made during the internship in CDS Lab link. The project was started as part of existed lab env - github link.