File size: 2,194 Bytes
862894c |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 |
---
library_name: skrl
tags:
- deep-reinforcement-learning
- reinforcement-learning
- skrl
model-index:
- name: PPO
results:
- metrics:
- type: mean_reward
value: 6550.07 +/- 619.16
name: Total reward (mean)
task:
type: reinforcement-learning
name: reinforcement-learning
dataset:
name: OmniIsaacGymEnvs-Humanoid
type: OmniIsaacGymEnvs-Humanoid
---
# OmniIsaacGymEnvs-Humanoid-PPO
Trained agent model for [NVIDIA Omniverse Isaac Gym](https://github.com/NVIDIA-Omniverse/OmniIsaacGymEnvs) environment
- **Task:** Humanoid
- **Agent:** [PPO](https://skrl.readthedocs.io/en/latest/modules/skrl.agents.ppo.html)
# Usage (with skrl)
```python
from skrl.utils.huggingface import download_model_from_huggingface
# assuming that there is an agent named `agent`
path = download_model_from_huggingface("skrl/OmniIsaacGymEnvs-Humanoid-PPO")
agent.load(path)
```
# Hyperparameters
```python
# https://skrl.readthedocs.io/en/latest/modules/skrl.agents.ppo.html#configuration-and-hyperparameters
cfg_ppo = PPO_DEFAULT_CONFIG.copy()
cfg_ppo["rollouts"] = 32 # memory_size
cfg_ppo["learning_epochs"] = 5
cfg_ppo["mini_batches"] = 4 # 32 * 4096 / 32768
cfg_ppo["discount_factor"] = 0.99
cfg_ppo["lambda"] = 0.95
cfg_ppo["learning_rate"] = 5e-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"] = 160
cfg_ppo["experiment"]["checkpoint_interval"] = 1600
```
|