
My Ai Sandbox
模拟·发行日期2026-09-30·Ainzlark Studio
全站在线第 6137 名
玩法介绍
这游戏干的事一句话:造一个3D世界,放一群AI进去,看着它们自己学会活下来。开局你从三个预设场景里挑一个(或者自己从零搭),设定环境和训练规则,然后训练启动——神经网络驱动的智能体种群在里头试错、进化,一代接一代。你的主要动作是搭建和调参,搭完之后基本就是观众,看它们跑。训练好的智能体能导出成ONNX格式(一种通用的AI模型交换格式,导出去能直接塞进你自己项目里加载运行),等于游戏就是你训练管线的沙盒调试器。 它跟一般沙盒最不一样的地方在0.2.0加的分阶段训练:训练可以从一个简单的环境起步,难度逐步升级,最终到达你设定的目标环境,而不是从头就在最难的场景里硬练;另外训练中途环境还能被重新排列,智能体不能死记位置。时间花在哪:前段是主动操作(搭环境、定规则),后段是被动观察(看它跑),一局没有固定时长,取决于你设多少代、环境多复杂。新手会卡在「怎么定规则、训到什么时候算够、ONNX导出去拿什么工具跑」这一串上,游戏不手把手带,首版内容也薄。
游戏资料
- 收费方式
- 买断制(需购买)
- 游戏类型
- 模拟
- 开发商
- Ainzlark Studio
- 发行商
- Ainzlark Studio
- 发行日期
- 2026-09-30
- 游戏特色
- 单人 · 自定义音量控制 · #category_playable_at_your_own_pace · 无需应对快速反应事件 · 家庭共享
- 支持语言
- 英语, 西班牙语 - 西班牙, 葡萄牙语 - 巴西, 法语, 德语, 简体中文, 日语, 俄语
热门用户自定义标签
Steam 官方介绍
Create, train, and modify artificial intelligence agents in 3D environments built by you.
In My AI Sandbox, agents are born knowing nothing and learn on their own, generation after generation, through neural networks, natural selection, mutation, and genetic crossover. No behavior is manually programmed.
You define the environment, objectives, risks, and rewards, and evolution takes care of the rest—testing thousands of generations until a behavior emerges that actually works.
KEY FEATURES
3D Environment Editor

Build your own environment from scratch. Place objects, obstacles, terrain, and risk or reward zones directly in the 3D scene using a dedicated editing mode with automatic saving after every change.
Agents Powered by Neural Networks

Configure sensors, hidden neurons, population size, mutation rate, crossover rate, and the duration of each generation. Each species evolves independently, with its own rewards and penalties.
Prebuilt and Custom Scenarios

Start with scenarios such as Evolutionary Survival, Predators and Prey, or Maze Navigation, or build your own environment with unique objects, obstacles, and creatures.
Parallel Training

Speed up evolution by running multiple environments at the same time and comparing populations side by side.
Real-Time Visualization

Track fitness charts, watch each agent’s neural network activate live, and analyze detailed statistics for each species while training is in progress.
Simulation Mode

Once training is complete, release your evolved agents to run freely without further training and simply observe the behaviors they have learned.
Save, Export, and Compare

Export trained agents, save entire projects, and compare different configurations to understand what actually worked.
系统需求
Windows
最低配置
- 操作系统
- Windows 10 or 11 64 Bit
- 处理器
- AMD Ryzen 3 3100 / Intel Core i3-10100
- 内存
- 8 GB RAM
- 显卡
- DirectX 11 compatible
- DirectX 版本
- 11
- 存储空间
- 需要 500 MB 可用空间
推荐配置
- 操作系统
- Windows 10 or 11 64 Bit
- 处理器
- AMD Ryzen 9 5900X / Intel Core i9-12900K
- 内存
- 16 GB RAM
- 显卡
- Any DirectX 11 compatible GPU
- DirectX 版本
- 11
- 存储空间
- 需要 1 GB 可用空间
相关游戏
在线人数走势历史日线 53 天
每日数据 (全部历史 53 天 · 每页 14 天 · 点 ▸ 展开当日每小时明细)
本页 2026-11-22 至 2026-11-09(第 1 / 4 页,第 1 页是最近两周)
| 日期 | 当日峰值 | 当日均值 | 当日最后 | 采样数 |
|---|---|---|---|---|
| ▸2026-11-22 | 0 | 0 | 0 | 23 |
| ▸2026-11-21 | 1 | 0 | 0 | 24 |
| ▸2026-11-20 | 1 | 0 | 0 | 24 |
| ▸2026-11-19 | 1 | 0 | 1 | 24 |
| ▸2026-11-18 | 1 | 0 | 0 | 24 |
| ▸2026-11-17 | 1 | 0 | 1 | 24 |
| 2026-11-16 | 0 | 0 | 0 | 24 |
| 2026-11-15 | 0 | 0 | 0 | 24 |
| 2026-11-14 | 0 | 0 | 0 | 24 |
| 2026-11-13 | 0 | 0 | 0 | 24 |
| 2026-11-12 | 0 | 0 | 0 | 24 |
| 2026-11-11 | 0 | 0 | 0 | 24 |
| 2026-11-10 | 0 | 0 | 0 | 23 |
| 2026-11-09 | 0 | 0 | 0 | 12 |
数据也可整站机读:/data/history/5238020.json 与 /data/top.json(供脚本、AI 直接读取,无需解析网页)。
appid 5238020 · 收录 53 天 · 历史日线 53 天
本页名词(点开看算法)
本页名词:当前在线·24H 峰值·较昨日·纪录墙·历史日线·当日峰值·当日均值·每小时线·Steam 实时榜·全部游戏榜·热门用户自定义标签·已停止跟踪
