deep-reinforcement-learning reinforcement-learning skrl

<!-- --- torch: -21.51 +/- 14.99 jax: -35.77 +/- 0.39 numpy: -8.89 +/- 10.3 --- -->

IsaacGymEnvs-FactoryTaskNutBoltScrew-PPO

Trained agent for NVIDIA Isaac Gym Preview environments.

Usage (with skrl)

Note: Visit the skrl Examples section to access the scripts.

Hyperparameters

Note: Undefined parameters keep their values by default.

# https://skrl.readthedocs.io/en/latest/api/agents/ppo.html#configuration-and-hyperparameters
cfg = PPO_DEFAULT_CONFIG.copy()
cfg["rollouts"] = 128  # memory_size
cfg["learning_epochs"] = 8
cfg["mini_batches"] = 32  # 128 * 128 / 512
cfg["discount_factor"] = 0.99
cfg["lambda"] = 0.95
cfg["learning_rate"] = 1e-4
cfg["random_timesteps"] = 0
cfg["learning_starts"] = 0
cfg["grad_norm_clip"] = 0
cfg["ratio_clip"] = 0.2
cfg["value_clip"] = 0.2
cfg["clip_predicted_values"] = True
cfg["entropy_loss_scale"] = 0.0
cfg["value_loss_scale"] = 1.0
cfg["kl_threshold"] = 0.016
cfg["rewards_shaper"] = None
cfg["state_preprocessor"] = RunningStandardScaler
cfg["state_preprocessor_kwargs"] = {"size": env.observation_space, "device": device}
cfg["value_preprocessor"] = RunningStandardScaler
cfg["value_preprocessor_kwargs"] = {"size": 1, "device": device}