Instructions to use Yunrufeng000/act_so100_4colors_540p with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- LeRobot
How to use Yunrufeng000/act_so100_4colors_540p with LeRobot:
# No code snippets available yet for this library. # To use this model, check the repository files and the library's documentation. # Want to help? PRs adding snippets are welcome at: # https://github.com/huggingface/huggingface.js
- Notebooks
- Google Colab
- Kaggle
File size: 1,510 Bytes
970a821 | 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 | ---
license: apache-2.0
library_name: lerobot
pipeline_tag: robotics
tags:
- lerobot
- act
- so100
- so-arm100
- robotics
---
# ACT · SO-100 · 四色抓取 (540×960)
基于 [LeRobot](https://github.com/huggingface/lerobot) 训练的 ACT 策略,用于 SO-100 (so_follower) 6 自由度机械臂的彩色物块抓取任务。
## ⚠️ 部署关键:输入分辨率
原始采集数据是 **1920×1080**,但本模型训练时把图像 **resize 到 540×960 (H×W)**。
**推理时必须把相机画面同样 resize 到 540×960**,否则会很慢且效果异常。
## 训练信息
| 项目 | 值 |
|---|---|
| 策略 | ACT (vision: ResNet18, chunk_size=100) |
| 机型 | SO-100 / so_follower,6 关节 |
| 相机 | 单路前置 (`observation.images.front`) |
| 输入分辨率 | 540×960 (从 1080p 缩放) |
| 训练步数 | 300,000 |
| Batch size | 8 |
| 最终 loss | 0.046 |
| 训练轮数 | ~82 epoch |
| fps | 30 |
| 硬件 | RTX 5080 (CUDA 12.8, torch 2.11) |
## 训练数据
合并自 8 个数据集(4 种颜色 × 2 轮采集),共 **77 episode / 29250 帧 / 4 个任务**:
- `Yunrufeng000/test_blue`, `test_blue1`
- `Yunrufeng000/test_green`, `test_green1`
- `Yunrufeng000/test_red`, `test_red1`
- `Yunrufeng000/test_yellow`, `test_yellow1`
## 使用
```python
from lerobot.policies.act.modeling_act import ACTPolicy
policy = ACTPolicy.from_pretrained("Yunrufeng000/act_so100_4colors_540p")
# 推理前记得把相机画面 resize 到 (540, 960)
```
|