[research] Look into ways to solve adjacent environments

[clay-arras]

description: this repository is based off of Car Racing, an OpenAI Gymnasium environment. the task is to look into existing policies/training strategies that can solve Car Racing, and try to apply them here

ideas:

• look into frame skipping, has proven effective in Atari environments
• look into taking the difference between frames to get the motion of the car
• customizing reward function and curriculum learning
• for the policy, we'll probably just use PPO from stable-baselines3, but if you're interested you can look into https://github.com/rl-tools/rl-tools

resources:

• https://notanymike.github.io/Solving-CarRacing/
• https://medium.com/@kartha.kishan/solving-openai-carracing-v0-using-image-processing-5e1005ee0cb
• https://github.com/wpiszlogin/driver_critic

[clay-arras]

Pranay