metadata

library_name: skrl
tags:
  - deep-reinforcement-learning
  - reinforcement-learning
  - skrl
model-index:
  - name: PPO
    results:
      - metrics:
          - type: mean_reward
            value: 9.1 +/- 0.05
            name: Total reward (mean)
        task:
          type: reinforcement-learning
          name: reinforcement-learning
        dataset:
          name: Isaac-Reach-Franka-v0
          type: Isaac-Reach-Franka-v0

IsaacOrbit-Isaac-Reach-Franka-v0-PPO

Trained agent model for NVIDIA Isaac Orbit environment

Task: Isaac-Reach-Franka-v0
Agent: PPO

Usage (with skrl)

from skrl.utils.huggingface import download_model_from_huggingface

# assuming that there is an agent named `agent`
path = download_model_from_huggingface("skrl/IsaacOrbit-Isaac-Reach-Franka-v0-PPO")
agent.load(path)

Hyperparameters

# https://skrl.readthedocs.io/en/latest/modules/skrl.agents.ppo.html#configuration-and-hyperparameters
cfg_ppo["rollouts"] = 16  # memory_size
cfg_ppo["learning_epochs"] = 8
cfg_ppo["mini_batches"] = 8  # 16 * 2048 / 4096
cfg_ppo["discount_factor"] = 0.99
cfg_ppo["lambda"] = 0.95
cfg_ppo["learning_rate"] = 3e-4
cfg_ppo["learning_rate_scheduler"] = KLAdaptiveRL
cfg_ppo["learning_rate_scheduler_kwargs"] = {"kl_threshold": 0.008}
cfg_ppo["random_timesteps"] = 0
cfg_ppo["learning_starts"] = 0
cfg_ppo["grad_norm_clip"] = 1.0
cfg_ppo["ratio_clip"] = 0.2
cfg_ppo["value_clip"] = 0.2
cfg_ppo["clip_predicted_values"] = True
cfg_ppo["entropy_loss_scale"] = 0.0
cfg_ppo["value_loss_scale"] = 2.0
cfg_ppo["kl_threshold"] = 0
cfg_ppo["rewards_shaper"] = lambda rewards, timestep, timesteps: rewards * 0.01
cfg_ppo["state_preprocessor"] = RunningStandardScaler
cfg_ppo["state_preprocessor_kwargs"] = {"size": env.observation_space, "device": device}
cfg_ppo["value_preprocessor"] = RunningStandardScaler
cfg_ppo["value_preprocessor_kwargs"] = {"size": 1, "device": device}
# logging to TensorBoard and writing checkpoints
cfg_ppo["experiment"]["write_interval"] = 40
cfg_ppo["experiment"]["checkpoint_interval"] = 400