File size: 1,871 Bytes
1bcfe99 | 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 | # SpaceMining PPO Agent
A PPO agent trained on the SpaceMining Gymnasium environment. This repository includes the final Stable-Baselines3 checkpoint, configuration, and evaluation metrics.
## Model Description
- Algorithm: PPO (Stable-Baselines3)
- Environment: SpaceMining (Gymnasium)
- Action Space: Box(3,) — thrust x, thrust y, mine toggle
- Observation Space: Box(53,) — agent state, nearby asteroids (up to 15), mothership relative position
## Quickstart
```python
from huggingface_hub import hf_hub_download
from stable_baselines3 import PPO
from space_mining import make_env
ckpt_path = hf_hub_download(repo_id="LUNDECHEN/space-mining-ppo", filename="final_model.zip")
model = PPO.load(ckpt_path)
env = make_env(render_mode='rgb_array')
obs, _ = env.reset()
for _ in range(300):
# SB3 `predict` may return `(action, state, *extras)` depending on version.
prediction = model.predict(obs, deterministic=True)
action = prediction[0] if isinstance(prediction, (tuple, list)) else prediction
obs, reward, terminated, truncated, info = env.step(action)
if terminated or truncated:
break
env.close()
```
## Training Configuration
- See `hyperparams.json` (algorithm hyperparameters)
- See `env_config.json` (environment parameters)
- See `training_args.json` (timesteps, device, versions)
## Evaluation
- See `evaluation.json`
| Metric | Value |
|---------------|-------|
| mean_reward | 1037.7470 |
| std_reward | 1449.5437 |
| episodes | 100 |
## Agent Behavior
## License
- MIT
## Authors
- Xinning Zhu (zhuxinning@shu.edu.cn)
- Lunde Chen (lundechen@shu.edu.cn)
## Training Details
- **Training Steps**: 5,000,000
- **Device**: cpu
- **Model Type**: best
- **GitHub Run**: [17421809264](https://github.com/reveurmichael/space_mining/actions/runs/17421809264)
|