This repository holds a sandbox driving simulation controllable via a network interface as a machine learning and data collection challenge.
From the root of the repository, run
pip install -r requirements.txt
pip install -e .
To run the game, you can use
python scripts/main.py
Launching main.py starts a race with a single car on the provided track. This track can be controlled either by keyboard (AWSD) or by a socket interface. An example of such interface is included in the code in example_data_collector.py.
The car sensing is available in two commodities: raycasts and images. These sensing snapshots are sent at 10 Hertz (i.e. 10 times a second). Due to this fact, correct recetion of snapshot messages has to be done regularly (See Server buffer saturation section).
A remote controller can be impemented using TCP socket connecting on localhost on port 7654. Different commands can be issued to the race simulation to control the car.
The car simulator expect specific commands to be sent. A command is a specific set of words or numbers. Spaces are used as elements separator in a single command. Semi-colon (;) as command separator.
push|release forward|right|left|back;
To simulate key press and control the car.
set position x,y,z;
To set the reset position at the (x,y,z) location. x,y,z are float numbers with english style dot separator (e,g. 3.1415, 2.5, 6.34234) themselves separated by a comma.
set rotation a;
to set the car orientation, with a being a float number in degrees
set speed v;
to set the starting speed after reset
reset;
to reset the car at the provided location.
set ray visible|hidden;
To toggle the ray sensor visibility. In direct play v can be used to toggle ray visibility
While implementing your own controller, make sure to regularly empty the socket buffer connected to the server by regularly calling NetworkDataCmdInterface.recv_msgs. Otherwise, while sending images, the server might fill the buffer and throw socket errors leading to a crash.
To run multiplayer, run the main.py file and click Multiplayer. Then enter the ip address (this can be defaulted to 'localhost') and the port (default: 25565). Click Create - Server and then click Join - Server.
For others to join, run the main.py, click Multiplayer and then click Join Server. Enter the public ip address and the port of the server. Click JOIN and then you're in. Unlimited people can join!
If you don't want to play on a server, run the main.py file and click Singleplayer.
W - Drive S - Brake A, D - Turn SPACE - Hand Brake ESCAPE - Pause Menu G - Respawn
We provide two docker containers to run the game: one that launches the usual graphic version (only runs on Linux host with X11), and one that runs the game on a headless server (will be used for model training).
The graphical container can be run using
docker compose run gui
and the headless one can be run using
docker compose run headless
This code is based on the repository https://github.com/mandaw2014/Rally