description.txt

BASIC REINFORCEMENT LEARNING IMPLEMENTATION FROM STABLE BASELINES


airsim_env: The AirSim-v0 OpenAI Gym env (install instructions at https://pypi.org/project/airsim/ and https://microsoft.github.io/AirSim/build_windows/)
Location - same as imitation_learning_basic.py


binaries: The AirSim binaries for AirSim-v0 (Windows, Linux). "Blocks.exe" must be open and needs to be detected for training AirSim-v0 (to access more Binary options, modify Documents/AirSim/settings.json accordingly. Description at https://microsoft.github.io/AirSim/settings/)


callbacks: save model periodically (checkpoints) and evaluate model periodically (evalautions)


experts: collection of .npz files. Note: AirSim-v0 expert data (generated from an optimal SAC policy). 1 -> 'simple' reward, 2 -> 'complex' reward (check __init__ of airsim_env_0.py in airsim_env for more info)


logs: log files for airsim experiments (check __init__ of airsim_env_0.py in airsim_env for more info)
logs/monitor: files from the Monitor wrapper

models: collection of all trained RL models


stable_baselines: The "stable_baselines" folder from the Stable Baselines 2.10 repo


tensorboard: Tensorboard logs according to env_id. Multiple runs of an experiment will result in multiple logs, please make sure to keep track of the runs (exp_id helps)


utils: to store the cmd line hyperparameters as a Dict