README.md

metadata

library_name: stable-baselines3
tags:
  - CarRacing-v0
  - deep-reinforcement-learning
  - reinforcement-learning
  - stable-baselines3
model-index:
  - name: RecurrentPPO
    results:
      - metrics:
          - type: mean_reward
            value: 880.39 +/- 31.90
            name: mean_reward
        task:
          type: reinforcement-learning
          name: reinforcement-learning
        dataset:
          name: CarRacing-v0
          type: CarRacing-v0

RecurrentPPO Agent playing CarRacing-v0

This is a trained model of a RecurrentPPO agent playing CarRacing-v0 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

# Download model and save it into the logs/ folder
python -m rl_zoo3.load_from_hub --algo ppo_lstm --env CarRacing-v0 -orga sb3 -f logs/
python enjoy.py --algo ppo_lstm --env CarRacing-v0  -f logs/

Training (with the RL Zoo)

python train.py --algo ppo_lstm --env CarRacing-v0 -f logs/
# Upload the model and generate video (when possible)
python -m rl_zoo3.push_to_hub --algo ppo_lstm --env CarRacing-v0 -f logs/ -orga sb3

Hyperparameters

OrderedDict([('batch_size', 128),
             ('clip_range', 0.2),
             ('ent_coef', 0.0),
             ('env_wrapper',
              [{'gym.wrappers.resize_observation.ResizeObservation': {'shape': 64}},
               {'gym.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', 4000000.0),
             ('normalize', "{'norm_obs': False, 'norm_reward': True}"),
             ('policy', 'CnnLstmPolicy'),
             ('policy_kwargs',
              'dict(log_std_init=-2, ortho_init=False, '
              'enable_critic_lstm=False, activation_fn=nn.GELU, '
              'lstm_hidden_size=128, )'),
             ('sde_sample_freq', 4),
             ('use_sde', True),
             ('vf_coef', 0.5),
             ('normalize_kwargs', {'norm_obs': False, 'norm_reward': False})])

Environment Arguments

{'verbose': 0}