Information About Environment

Checkout environment.md for details.

Environment Setup

Download Simstar from the link:

https://drive.google.com/open?id=1Gp-XXnOX9dbDcfqFJNJ4UtZqo9sWqjUg

Version 2.0.1 Update [20 Jun 2021]

Download V.2.01 from the link above. Please reinstall simstar PythonAPI for the update to be compatible.

Change Log

• New tracks are added: Circular and Straight
• Flag marks are added ever 200m of the tracks
• evaluate.py is updated to examplify a race with trained agents

Windows

Just click on Simstar.exe and Simstar is ready.

Linux

Requirements

Install Vulkan Libraries

sudo apt-get install vulkan-utils

cd Simstar

chmod 777 -R *

./Simstar.sh

Requirements

Python Package Requirements

Option 1: Install using Anaconda

Create a new environment using anaconda.

conda env create --file environment.yml

conda activate simstar-env

Option 2: Install using pip

Install required python libraries from requirements.txt by

pip install -r requirements.txt

Pytorch Version

The final evaluation will be using pytorch version 1.5 and CUDA version 10.2.

Install Python API

  cd PythonAPI

  python setup.py install --user

Installation Test

There are multiple stages that needs to be checked.

1. Test Simstar Executable

Open the simstar executable, allow for networking if asked.

2. Test Environment Setup

cd sac_example

python example_experiment.py

Optional Test

To test a closed loop training with Pytorch, you can run the example SAC agent from examples folder.

cd sac_example

python train_example.py