---
title: hands-on-modern-rl-unity-environments
canonical_url: "https://www.modelscope.cn/datasets/walkinglab/hands-on-modern-rl-unity-environments"
md_url: "https://www.modelscope.cn/datasets/walkinglab/hands-on-modern-rl-unity-environments.md"
repository: walkinglab/hands-on-modern-rl-unity-environments
chinese_name: "Hands-On Modern RL · Unity 场景包大全"
last_updated: 2026-08-17
license: other
downloads: 575
stars: 0
---

# hands-on-modern-rl-unity-environments

> hands-on-modern-rl-unity-environments - walkinglab 在 ModelScope 开源的数据集。Reusable Linux Unity environment bundles for WalkingLab Hands-On Modern RL online training demos.

walkinglab/hands-on-modern-rl-unity-environments 是 ModelScope 魔搭社区上的数据集，采用 other 许可。

- **Repository**: walkinglab/hands-on-modern-rl-unity-environments
- **License**: other
- **Downloads**: 575
- **Stars**: 0
- **Last updated**: 2026-08-17

Source: https://www.modelscope.cn/datasets/walkinglab/hands-on-modern-rl-unity-environments

---

# WalkingLab × Hands-On Modern RL · Unity Environment Packs

Reusable **Linux x86_64 Unity environments** for the online training demos in [Hands-On Modern RL](https://github.com/walkinglabs/hands-on-modern-rl). The bundles are kept in one versioned Dataset so every WalkingLab Studio can use the same checked assets instead of downloading them from several external hosts.

这是 [WalkingLab](https://modelscope.cn/organization/walkinglab) 为 [《动手学现代强化学习》](https://walkinglabs.github.io/hands-on-modern-rl/)维护的 Unity Linux 场景包集合。它集中保存在线训练实验使用的可执行场景、来源、版本和 SHA-256，供多个创空间复用。

## Included packs

| Pack | File | Scenes | Size | SHA-256 |
| --- | --- | ---: | ---: | --- |
| Huggy · Fetch the Stick | `linux/huggy/Huggy.zip` | 1 | 39,214,997 B | `6b35692b1d867f74fdf8987a911700e06ff24d40b95b935460ccd175e3712d28` |
| Unity ML-Agents 1.1.0 Startup | `linux/ml-agents-1.1.0/Startup.zip` | 18 | 72,623,449 B | `80e2322215fb7ff5c192e34bd67d63edc80d8cf24e66f8af858010b84a250a5d` |

The Startup build contains the launcher plus 17 ML-Agents example scenes: Basic, 3D Ball, 3D Ball Hard, Crawler, Dungeon Escape, Food Collector, GridWorld, Hallway, Push Block, Pyramids, Soccer Twos, Sorter, Strikers vs Goalie, Visual Food Collector, Wall Jump, Walker, and Worm.

完整机器可读列表见 [`manifest.json`](./manifest.json)，文件校验值见 [`checksums.sha256`](./checksums.sha256)。

## Real environment captures

All preview files below are compressed captures of the actual Unity environments, stored locally in this Dataset rather than hotlinked at runtime.

| Huggy | Basic | 3D Ball |
| --- | --- | --- |
| ![Huggy](./previews/huggy.webp) | ![Basic](./previews/basic.webp) | ![3D Ball](./previews/3dball.webp) |

| Food Collector | Walker |
| --- | --- |
| ![Food Collector](./previews/food-collector.webp) | ![Walker](./previews/walker.webp) |

## Download with ModelScope

```bash
pip install modelscope

# Huggy only
modelscope download --dataset walkinglab/hands-on-modern-rl-unity-environments \
  linux/huggy/Huggy.zip --local_dir ./unity-environments

# Official multi-scene ML-Agents build only
modelscope download --dataset walkinglab/hands-on-modern-rl-unity-environments \
  linux/ml-agents-1.1.0/Startup.zip --local_dir ./unity-environments
```

Download the complete Dataset in Python:

```python
from modelscope import dataset_snapshot_download

dataset_dir = dataset_snapshot_download(
    "walkinglab/hands-on-modern-rl-unity-environments"
)
print(dataset_dir)
```

## Launch a scene

```bash
unzip -q linux/ml-agents-1.1.0/Startup.zip
chmod +x Startup/Startup.x86_64

./Startup/Startup.x86_64 \
  --mlagents-scene-name \
  Assets/ML-Agents/Examples/Walker/Scenes/Walker.unity
```

Launch Huggy:

```bash
unzip -q linux/huggy/Huggy.zip
chmod +x Huggy/Huggy.x86_64
./Huggy/Huggy.x86_64
```

For a headless server, run the executable with Xvfb when rendered frames are required. The companion [Unity ML-Agents xGPU Studio](https://modelscope.cn/studios/walkinglab/hands-on-modern-rl-experiment11-unity-mlagents) already configures Xvfb, training logs, live frames, and replay export.

## Sources and usage rights

- `Huggy.zip` is an unmodified copy of the build published by [huggingface/Huggy](https://github.com/huggingface/Huggy). The upstream repository does not currently publish a license file; review the upstream terms before redistribution or commercial use.
- `Startup.zip` is an unmodified official build from [Unity ML-Agents](https://github.com/Unity-Technologies/ml-agents). ML-Agents source is distributed under Apache-2.0; Unity runtime and packaged assets can have additional Unity terms.
- The Huggy preview is the real capture referenced by the official Huggy README. Basic, 3D Ball, Food Collector, and Walker previews are real captures from the Unity ML-Agents Release 20 documentation assets.

The Dataset is marked `other` because the two upstream archives do not share one uniform license. WalkingLab does not change or relicense either archive.