pableitorr/ppo-CarRacing-v2

PPO Agent playing CarRacing-v2

This is a trained model of a PPO agent playing CarRacing-v2 using the stable-baselines3 library and the RL Zoo.

The RL Zoo is a training framework for Stable Baselines3 reinforcement learning agents, with hyperparameter optimization and pre-trained agents included.

Usage (with SB3 RL Zoo)

RL Zoo: https://github.com/DLR-RM/rl-baselines3-zoo
SB3: https://github.com/DLR-RM/stable-baselines3
SB3 Contrib: https://github.com/Stable-Baselines-Team/stable-baselines3-contrib

Install the RL Zoo (with SB3 and SB3-Contrib):

pip install rl_zoo3

# Download model and save it into the logs/ folder
python -m rl_zoo3.load_from_hub --algo ppo --env CarRacing-v2 -orga pableitorr -f logs/
python -m rl_zoo3.enjoy --algo ppo --env CarRacing-v2  -f logs/

If you installed the RL Zoo3 via pip (pip install rl_zoo3), from anywhere you can do:

python -m rl_zoo3.load_from_hub --algo ppo --env CarRacing-v2 -orga pableitorr -f logs/
python -m rl_zoo3.enjoy --algo ppo --env CarRacing-v2  -f logs/

Training (with the RL Zoo)

python -m rl_zoo3.train --algo ppo --env CarRacing-v2 -f logs/
# Upload the model and generate video (when possible)
python -m rl_zoo3.push_to_hub --algo ppo --env CarRacing-v2 -f logs/ -orga pableitorr

Hyperparameters

OrderedDict([('batch_size', 128),
             ('clip_range', 0.2),
             ('ent_coef', 0.0),
             ('env_wrapper',
              [{'rl_zoo3.wrappers.FrameSkip': {'skip': 2}},
               {'gymnasium.wrappers.resize_observation.ResizeObservation': {'shape': 64}},
               {'gymnasium.wrappers.gray_scale_observation.GrayScaleObservation': {'keep_dim': True}}]),
             ('frame_stack', 2),
             ('gae_lambda', 0.95),
             ('gamma', 0.99),
             ('learning_rate', 'lin_1e-4'),
             ('max_grad_norm', 0.5),
             ('n_envs', 8),
             ('n_epochs', 10),
             ('n_steps', 512),
             ('n_timesteps', 1000000),
             ('normalize', "{'norm_obs': False, 'norm_reward': True}"),
             ('policy', 'CnnPolicy'),
             ('policy_kwargs',
              'dict(log_std_init=-2, ortho_init=False, activation_fn=nn.GELU, '
              'net_arch=dict(pi=[256], vf=[256]), )'),
             ('sde_sample_freq', 4),
             ('use_sde', True),
             ('vf_coef', 0.5),
             ('normalize_kwargs', {'norm_obs': False, 'norm_reward': False})])

Environment Arguments

{'render_mode': 'rgb_array'}