// the find
Sentdex/Carla-RL
Reinforcement Learning codebase for self-driving car in Carla
A reinforcement learning setup for training self-driving agents inside the Carla simulator. It's a research/learning project from Sentdex (the Python tutorial guy), aimed at people who want a working starting point for RL with Carla rather than building from scratch.
Clean separation between agent, trainer, and model code makes it readable; settings.py centralizes hyperparameters so you're not hunting through files to change them; Tensorboard integration is already wired up, which saves real setup time; the play.py script lets you manually drive and observe the environment before training, which is genuinely useful for debugging reward functions.
Last touched in May 2020 — Carla has had multiple breaking API changes since then and this almost certainly won't run against a current Carla install without significant patching; the README is three lines, so if you hit an issue you're on your own; reward function and environment logic live in sources/carla.py with no documentation of design choices, making it hard to know what to tune; no pretrained weights or training curves, so you can't verify the code actually converges to anything useful before committing hours of GPU time.