Skip to content

Commit c7a209c

Browse files
lxsy-xcylxsy-xcy
committed
fix
1 parent 9df6a55 commit c7a209c

4 files changed

Lines changed: 163 additions & 21 deletions

File tree

‎README.md‎

Lines changed: 79 additions & 9 deletions
Original file line numberDiff line numberDiff line change
@@ -137,7 +137,7 @@ See the [LIBERO simulation evaluation guide](eval/libero/README.md).
137137

138138
### 🦾 Using model-server on real hardware, using SO-101 as the example
139139

140-
See the [SO-101 deployment guide](eval/lerobot_so101/README.md).
140+
See the [SO-101 deployment guide](eval/lerobot_so101/README.md).
141141

142142
---
143143

@@ -167,14 +167,64 @@ For SmolVLA preprocessing, we follow the official default setting: images are fi
167167

168168
This section lists converted GGUF models that can be used directly with `model-server` for smoke tests and quick starts. For your own real-world scenarios, we recommend using [`hf2gguf`](tools/hf2gguf/README.md) to generate your own GGUF models. Different components can also use different precisions; in practice, the best precision choice is often component-specific. In our examples, the state projector always stays in f32, while the other GGUF files follow the listed precision. You can mix and match them to explore better accuracy/performance tradeoffs.
169169

170-
| Model | Benchmark | Precision | Link |
171-
| ------- | --------- | --------- | ----------------------------------------------------------------------- |
172-
| SmolVLA | SO-101 | bf16 | [smolvla-so101-bf16](https://huggingface.co/rrobottt/smolvla-so101-bf16) |
173-
| SmolVLA | SO-101 | f16 | [smolvla-so101-fp16](https://huggingface.co/rrobottt/smolvla-so101-fp16) |
174-
| SmolVLA | SO-101 | f32 | [smolvla-so101-fp32](https://huggingface.co/rrobottt/smolvla-so101-fp32) |
175-
| pi0 | LIBERO | bf16 | [pi-libero-bf16](https://huggingface.co/rrobottt/pi-libero-bf16) |
176-
| pi0 | LIBERO | f16 | [pi0-libero-f16](https://huggingface.co/rrobottt/pi0-libero-f16) |
177-
| pi0 | LIBERO | f32 | [pi0-libero-f32](https://huggingface.co/rrobottt/pi0-libero-f32) |
170+
<table>
171+
<thead>
172+
<tr>
173+
<th>Model</th>
174+
<th>Benchmark</th>
175+
<th>Original Python Model</th>
176+
<th>Precision</th>
177+
<th>GGUF Model</th>
178+
</tr>
179+
</thead>
180+
<tbody>
181+
<tr>
182+
<td rowspan="3">SmolVLA</td>
183+
<td rowspan="3">SO-101</td>
184+
<td rowspan="3"><code>smolvla_450m_grasp_50k</code> (local checkpoint)</td>
185+
<td>bf16</td>
186+
<td><a href="https://huggingface.co/rrobottt/smolvla-so101-bf16">smolvla-so101-bf16</a></td>
187+
</tr>
188+
<tr>
189+
<td>f16</td>
190+
<td><a href="https://huggingface.co/rrobottt/smolvla-so101-fp16">smolvla-so101-fp16</a></td>
191+
</tr>
192+
<tr>
193+
<td>f32</td>
194+
<td><a href="https://huggingface.co/rrobottt/smolvla-so101-fp32">smolvla-so101-fp32</a></td>
195+
</tr>
196+
<tr>
197+
<td rowspan="3">SmolVLA</td>
198+
<td rowspan="3">LIBERO</td>
199+
<td rowspan="3"><a href="https://huggingface.co/lerobot/smolvla_libero">lerobot/smolvla_libero</a></td>
200+
<td>bf16</td>
201+
<td><a href="https://huggingface.co/rrobottt/smolvla-libero-bf16">smolvla-libero-bf16</a></td>
202+
</tr>
203+
<tr>
204+
<td>f16</td>
205+
<td><a href="https://huggingface.co/rrobottt/smolvla-libero-fp16">smolvla-libero-fp16</a></td>
206+
</tr>
207+
<tr>
208+
<td>f32</td>
209+
<td><a href="https://huggingface.co/rrobottt/smolvla-libero-fp32">smolvla-libero-fp32</a></td>
210+
</tr>
211+
<tr>
212+
<td rowspan="3">pi0</td>
213+
<td rowspan="3">LIBERO</td>
214+
<td rowspan="3"><a href="https://huggingface.co/lerobot/pi0_libero_finetuned_v044">lerobot/pi0_libero_finetuned_v044</a></td>
215+
<td>bf16</td>
216+
<td><a href="https://huggingface.co/rrobottt/pi-libero-bf16">pi-libero-bf16</a></td>
217+
</tr>
218+
<tr>
219+
<td>f16</td>
220+
<td><a href="https://huggingface.co/rrobottt/pi0-libero-f16">pi0-libero-f16</a></td>
221+
</tr>
222+
<tr>
223+
<td>f32</td>
224+
<td><a href="https://huggingface.co/rrobottt/pi0-libero-f32">pi0-libero-f32</a></td>
225+
</tr>
226+
</tbody>
227+
</table>
178228

179229
---
180230

@@ -233,6 +283,26 @@ Issues and PRs are welcome. For larger model-architecture changes, protocol chan
233283

234284
---
235285

286+
## 📝 Citation
287+
288+
Robot.cpp is developed by the VIPL-VSU group, Key Laboratory of AI Safety, Institute of Computing Technology, Chinese Academy of Sciences (CAS), Beijing 100190, China.
289+
290+
**Project Lead:** Chuyan Xiong ([@lxsy-xcy](https://github.com/lxsy-xcy))
291+
**Corresponding Author:** Ruiping Wang ([wangruiping@ict.ac.cn](mailto:wangruiping@ict.ac.cn))
292+
293+
If you use Robot.cpp in your project, please cite the GitHub repository to acknowledge its ongoing development and contributors:
294+
295+
```bibtex
296+
@misc{robotcpp2026,
297+
author = {Chuyan Xiong, Sutong Wang, Jie Huang, Hong Chang, Ruiping Wang},
298+
title = {Robot.cpp: Run Your Robot Model Easily on Any Device},
299+
howpublished = {\url{https://github.com/Robot-cpp/robot.cpp}},
300+
year = {2026}
301+
}
302+
```
303+
304+
---
305+
236306
## 📄 License
237307

238308
Robot.cpp source code is released under the Apache License, Version 2.0. See [LICENSE](LICENSE) for the full license text.

‎README_ZH.md‎

Lines changed: 78 additions & 8 deletions
Original file line numberDiff line numberDiff line change
@@ -167,14 +167,64 @@ bash robot_client/shell/cpp_client_example.sh
167167

168168
这里整理一些已经转换好的 GGUF 模型，可以直接配合 `model-server` 做smoke test，以方便quick start！但针对自己的实际场景，我们推荐使用[hf2gguf](tools/hf2gguf/README_ZH.md)l来生成自己的GGUF model！并且对于不同的部分，您还可以自定义不同的精度，来实现不同部分的精度组合（事实上，不同部分的最优精度通常是不同的），我们的例子中，state proj始终保持f32精度，其他的gguf随着precision精度变化而变化，您可以自行组合，探索更好更高效的性能tradeoff！
169169

170-
| Model | Benchmark | Precision | Link |
171-
| ------- | --------- | --------- | ----------------------------------------------------------------------- |
172-
| SmolVLA | SO-101 | bf16 | [smolvla-so101-bf16](https://huggingface.co/rrobottt/smolvla-so101-bf16) |
173-
| SmolVLA | SO-101 | f16 | [smolvla-so101-fp16](https://huggingface.co/rrobottt/smolvla-so101-fp16) |
174-
| SmolVLA | SO-101 | f32 | [smolvla-so101-fp32](https://huggingface.co/rrobottt/smolvla-so101-fp32) |
175-
| pi0 | LIBERO | bf16 | [pi-libero-bf16](https://huggingface.co/rrobottt/pi-libero-bf16) |
176-
| pi0 | LIBERO | f16 | [pi0-libero-f16](https://huggingface.co/rrobottt/pi0-libero-f16) |
177-
| pi0 | LIBERO | f32 | [pi0-libero-f32](https://huggingface.co/rrobottt/pi0-libero-f32) |
170+
<table>
171+
<thead>
172+
<tr>
173+
<th>模型</th>
174+
<th>Benchmark</th>
175+
<th>原始 Python 模型</th>
176+
<th>精度</th>
177+
<th>GGUF 模型</th>
178+
</tr>
179+
</thead>
180+
<tbody>
181+
<tr>
182+
<td rowspan="3">SmolVLA</td>
183+
<td rowspan="3">SO-101</td>
184+
<td rowspan="3"><code>smolvla_450m_grasp_50k</code>（本地 checkpoint）</td>
185+
<td>bf16</td>
186+
<td><a href="https://huggingface.co/rrobottt/smolvla-so101-bf16">smolvla-so101-bf16</a></td>
187+
</tr>
188+
<tr>
189+
<td>f16</td>
190+
<td><a href="https://huggingface.co/rrobottt/smolvla-so101-fp16">smolvla-so101-fp16</a></td>
191+
</tr>
192+
<tr>
193+
<td>f32</td>
194+
<td><a href="https://huggingface.co/rrobottt/smolvla-so101-fp32">smolvla-so101-fp32</a></td>
195+
</tr>
196+
<tr>
197+
<td rowspan="3">SmolVLA</td>
198+
<td rowspan="3">LIBERO</td>
199+
<td rowspan="3"><a href="https://huggingface.co/lerobot/smolvla_libero">lerobot/smolvla_libero</a></td>
200+
<td>bf16</td>
201+
<td><a href="https://huggingface.co/rrobottt/smolvla-libero-bf16">smolvla-libero-bf16</a></td>
202+
</tr>
203+
<tr>
204+
<td>f16</td>
205+
<td><a href="https://huggingface.co/rrobottt/smolvla-libero-fp16">smolvla-libero-fp16</a></td>
206+
</tr>
207+
<tr>
208+
<td>f32</td>
209+
<td><a href="https://huggingface.co/rrobottt/smolvla-libero-fp32">smolvla-libero-fp32</a></td>
210+
</tr>
211+
<tr>
212+
<td rowspan="3">pi0</td>
213+
<td rowspan="3">LIBERO</td>
214+
<td rowspan="3"><a href="https://huggingface.co/lerobot/pi0_libero_finetuned_v044">lerobot/pi0_libero_finetuned_v044</a></td>
215+
<td>bf16</td>
216+
<td><a href="https://huggingface.co/rrobottt/pi-libero-bf16">pi-libero-bf16</a></td>
217+
</tr>
218+
<tr>
219+
<td>f16</td>
220+
<td><a href="https://huggingface.co/rrobottt/pi0-libero-f16">pi0-libero-f16</a></td>
221+
</tr>
222+
<tr>
223+
<td>f32</td>
224+
<td><a href="https://huggingface.co/rrobottt/pi0-libero-f32">pi0-libero-f32</a></td>
225+
</tr>
226+
</tbody>
227+
</table>
178228

179229
---
180230

@@ -233,6 +283,26 @@ robot.cpp 欢迎社区贡献新的模型 runtime、平台适配、评测流程
233283

234284
---
235285

286+
## 📝 引用
287+
288+
Robot.cpp 由中国科学院计算技术研究所智能算法安全全国重点实验室VIPL-VSU 课题组（中国北京，100190）开发。
289+
290+
**项目负责人：** Chuyan Xiong（[@lxsy-xcy](https://github.com/lxsy-xcy)）
291+
**通讯作者：** Ruiping Wang（[wangruiping@ict.ac.cn](mailto:wangruiping@ict.ac.cn)）
292+
293+
如果你在项目中使用了 Robot.cpp，请引用本 GitHub 仓库，以感谢项目持续开发过程中所有贡献者的工作：
294+
295+
```bibtex
296+
@misc{robotcpp2026,
297+
author = {Chuyan Xiong, Sutong Wang, Jie Huang, Hong Chang, Ruiping Wang},
298+
title = {Robot.cpp: Run Your Robot Model Easily on Any Device},
299+
howpublished = {\url{https://github.com/Robot-cpp/robot.cpp}},
300+
year = {2026}
301+
}
302+
```
303+
304+
---
305+
236306
## 📄 License
237307

238308
robot.cpp 源码使用 Apache License, Version 2.0 开源。完整协议文本见[LICENSE](LICENSE)。

‎eval/libero/README.md‎

Lines changed: 3 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -162,7 +162,8 @@ python -m eval.libero.runners.run_lerobot \
162162
The runner writes `stdout.log`, `stderr.log`, `baseline_run.json`, and
163163
LeRobot's `eval_info.json` under `eval/results/lerobot-baseline-*`.
164164

165-
For SmolVLA, use `HuggingFaceVLA/smolvla_libero` (its inputs match the GGUF).
165+
For SmolVLA, use [`lerobot/smolvla_libero`](https://huggingface.co/lerobot/smolvla_libero)
166+
(its inputs match the GGUF).
166167

167168
## C++ Policy Request Semantics
168169

@@ -228,4 +229,4 @@ GPU rendering is roughly an order of magnitude faster than OSMesa software
228229
rendering (~15 ms/step vs ~120 ms/step in our tests), which matters a lot for
229230
full-suite rollouts. Only fall back to `MUJOCO_GL=osmesa`
230231
(`apt-get install -y libosmesa6`, `PYOPENGL_PLATFORM=osmesa`) when no GPU
231-
rendering is available.
232+
rendering is available.

‎eval/libero/README_ZH.md‎

Lines changed: 3 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -153,7 +153,8 @@ python -m eval.libero.runners.run_lerobot \
153153
runner 会把 `stdout.log`、`stderr.log`、`baseline_run.json` 和 LeRobot 的
154154
`eval_info.json` 写到 `eval/results/lerobot-baseline-*` 目录下。
155155

156-
SmolVLA 请用 `HuggingFaceVLA/smolvla_libero`（其输入布局与 GGUF 一致）。
156+
SmolVLA 请使用 [`lerobot/smolvla_libero`](https://huggingface.co/lerobot/smolvla_libero)
157+
（其输入布局与 GGUF 一致）。
157158

158159
## C++ Policy 请求语义
159160

@@ -206,4 +207,4 @@ sudo tee /usr/share/glvnd/egl_vendor.d/10_nvidia.json >/dev/null <<'EOF'
206207
EOF
207208
# 然后：
208209
export MUJOCO_GL=egl MUJOCO_EGL_DEVICE_ID=0 # 与 CUDA_VISIBLE_DEVICES 对应
209-
```
210+
```

0 commit comments

Comments
 (0)