/supertuxkart_rl

use RL to train penguins to drive

Primary LanguagePython

CS 394R Final Project: Deep Kart Racing

Group members: Ann Yue and Brady Zhou.
This repository contains the codes for our CS 394R final project.

All experiment logs, hyperparameters and videos can be found at link.

Installation

Give it a minute or 10.

source install.sh

Usage

The current configuration is for 8+ core machine with GPU (GTX 1080ti or better).
Check spc/train.py for more details, but this is a minimal example of a policy trained with PPO that trains within an hour.

python3 -m spc.train --algorithm ppo \
  --gamma 0.9 \
  --lr 5e-4 \
  --batch_size 256 \
  --frame_skip 1 \
  --iterations 10 \
  --max_frames 10000