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
```