README.md

---
library_name: skrl
tags:
- deep-reinforcement-learning
- reinforcement-learning
- skrl
model-index:
- name: PPO
  results:
  - metrics:
    - type: mean_reward
      value: 298.89 +/- 27.4
      name: Total reward (mean)
    task:
      type: reinforcement-learning
      name: reinforcement-learning
    dataset:
      name: IsaacGymEnvs-BallBalance
      type: IsaacGymEnvs-BallBalance
---

<!-- ---
torch: 298.89 +/- 27.4
jax: 256.32 +/- 12.84
numpy: 240.59 +/- 19.15
--- -->

# IsaacGymEnvs-BallBalance-PPO

Trained agent for [NVIDIA Isaac Gym Preview](https://github.com/NVIDIA-Omniverse/IsaacGymEnvs) environments.

- **Task:** BallBalance
- **Agent:** [PPO](https://skrl.readthedocs.io/en/latest/api/agents/ppo.html)

# Usage (with skrl)

Note: Visit the skrl [Examples](https://skrl.readthedocs.io/en/latest/intro/examples.html) section to access the scripts.

* PyTorch

    ```python
    from skrl.utils.huggingface import download_model_from_huggingface

    # assuming that there is an agent named `agent`
    path = download_model_from_huggingface("skrl/IsaacGymEnvs-BallBalance-PPO", filename="agent.pt")
    agent.load(path)
    ```

* JAX

    ```python
    from skrl.utils.huggingface import download_model_from_huggingface

    # assuming that there is an agent named `agent`
    path = download_model_from_huggingface("skrl/IsaacGymEnvs-BallBalance-PPO", filename="agent.pickle")
    agent.load(path)
    ```

# Hyperparameters

Note: Undefined parameters keep their values by default.

```python
# https://skrl.readthedocs.io/en/latest/api/agents/ppo.html#configuration-and-hyperparameters
cfg = PPO_DEFAULT_CONFIG.copy()
cfg["rollouts"] = 16  # memory_size
cfg["learning_epochs"] = 8
cfg["mini_batches"] = 8  # 16 * 4096 / 8192
cfg["discount_factor"] = 0.99
cfg["lambda"] = 0.95
cfg["learning_rate"] = 3e-4
cfg["learning_rate_scheduler"] = KLAdaptiveRL
cfg["learning_rate_scheduler_kwargs"] = {"kl_threshold": 0.008}
cfg["random_timesteps"] = 0
cfg["learning_starts"] = 0
cfg["grad_norm_clip"] = 1.0
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"] = 2.0
cfg["kl_threshold"] = 0
cfg["rewards_shaper"] = lambda rewards, timestep, timesteps: rewards * 0.1
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}
```