Future directions
Ideas that are recorded but not scheduled and not started. Nothing in this file authorizes code, an asset, an upload, or a deployment; each part says what would have to be decided before it could.
Read [OPEN-WORK.md](OPEN-WORK.md) instead for work that is already underway or already owed. The line between the two files is deliberate: open work has a place it belongs in the repository today, and everything here does not yet.
Merged 2026-08-31 from docs/SPATIAL-GALLERY-AUTHORING.md and docs/FUTURE-PLAN.zh.md, which this file replaces.
| Part | Subject | State |
|---|---|---|
| 1 | Spatial gallery authoring: Art Lab to Inception Space | architecture proposal; no runtime schema |
| 2 | Technology & Biology Research Station — a drafted room and its LOE | draft room, never built |
| 3 | Capture and modelling tooling research (motion capture, photogrammetry) | research notes; nothing purchased |
| 4 | [FUTURE] items recorded in the product direction | design-only, listed here so they are findable |
Part 1 — Spatial gallery authoring: Art Lab to Inception Space
Spatial gallery authoring: Art Lab to Inception Space
Status: future architecture and product plan. This is not a runtime schema yet and does not authorize accepting unvalidated public uploads.
Product idea
The future gallery should be able to branch like a small maze, but the maze is not an adult-designed container that children merely decorate. Designing the space is part of the art class.
Lucas Visual Art Lab can add a Space Design tab beside its existing painting tools. A student arranges bounded rooms, corridors, doors, exhibit surfaces, and approved objects. Art Lab exports a declarative spatial design; Inception Space validates and imports it. Inception Space remains responsible for the real 3D runtime, collision, accessibility, privacy, performance, and publication state.
Art Lab · Space Design tab
│
│ export reviewed declarative design
▼
.is-space.json
│
│ validate, preview, approve, import
▼
Inception Space room/world document
The seam is intentional. Art Lab is the authoring environment; Inception Space is the publishing and multiplayer environment. Neither product executes code from the other.
What children design
The first version should offer meaningful composition rather than unrestricted 3D modelling:
- choose from approved room modules: cube, long gallery, L room, junction, small dome, and courtyard;
- connect modules with doors and corridors on a visible grid;
- name chapters and assign a topic, artist, or class collection to each zone;
- place exhibit walls, frames, plinths, labels, portals, and approved furniture;
- assign 2D artworks, Live Paintings, and reviewed 3D objects to surfaces;
- choose a route emphasis: chronological, thematic, compare-and-contrast, or free discovery;
- preview first-person and third-person scale before export;
- invite an AI critique that proposes a structured change and waits for the student to approve or reject it.
Children do not author JavaScript, shaders, arbitrary model scripts, network links, access-control rules, or collision code.
A small declarative exchange format
Use a versioned file such as .is-space.json. The first contract should stay smaller than the full Inception Space room document:
{
"format": "lucas-is-space/1",
"id": "class-gallery-draft",
"title": "Questions and Answers",
"units": "metres",
"modules": [
{ "id": "entry", "kind": "long-gallery", "at": [0, 0], "turn": 0 },
{ "id": "students", "kind": "cube", "at": [0, 12], "turn": 0 }
],
"connections": [
{ "from": "entry:north", "to": "students:south", "kind": "door" }
],
"chapters": [
{ "module": "entry", "title": "The Artist's Questions" },
{ "module": "students", "title": "Young Artists Respond" }
],
"placements": [
{ "key": "work-1", "module": "entry", "surface": "west", "asset": "approved-artwork-id" }
]
}
The final schema needs explicit numeric bounds, allowlisted enums, stable placement keys, declared format/version, and no executable strings. Asset ids refer to reviewed libraries; the archive does not smuggle arbitrary bytes into the public museum.
Validation before import
Art Lab should validate continuously, and Inception Space must validate again without trusting the export.
- Every module is inside the size and count budget.
- Every module is reachable from the entry.
- Every connected door has enough width and vertical clearance.
- No rooms, walls, doors, or exhibit boards overlap illegally.
- There is always a route back; a child cannot accidentally trap a visitor.
- Accessible routes avoid mandatory jumps, flight, tiny gaps, and confusing one-way passages.
- Artwork faces retain aspect ratio and stay within their mounting surfaces.
- Every asset exists in an approved library and is usable by the author.
- Performance budgets cap visible artworks, texture memory, active Live Paintings, dynamic lights, geometry, and adjacent streamed modules.
- Privacy and publication state are assigned by Inception Space accounts, not accepted from the portable design file.
A failed import should list fixable design problems in child-readable language and preserve the original file. It must never silently repair and publish a different space.
Maze without disorientation
The experience can feel maze-like without becoming hostile:
- each junction carries a chapter name and a visible return direction;
- a small optional map shows visited rooms, not private or locked rooms;
- route color and architectural motifs repeat by chapter;
- dead ends contain a deliberate exhibit or reflection point;
- a Return to entrance action is always available;
- teacher preview can show shortest paths, unreachable zones, crowding, and content density before approval.
The first release should cap a student design at roughly 8 modules, 12 connections, and 40 exhibit placements. Raise those limits only after device performance and classroom usability are measured.
Workflow
- A student opens Art Lab's Space Design tab and starts from a teacher-approved template or an empty bounded grid.
- The student builds, previews, and asks for critique.
- Art Lab records student intent, AI proposal, student decision, and revision.
- The student exports
.is-space.json. - Inception Space imports into a personal draft and runs its own validation.
- The student and educator preview the real 3D result.
- The student explicitly shares to the class; a separate curated process is required for any public snapshot.
Round-trip editing is a later phase. Version one is Art Lab export → Inception Space import. If Inception Space later exports back, it must retain unknown future-safe metadata without treating it as runtime behavior.
Delivery phases and high-level LOE
| Phase | Outcome | LOE |
|---|---|---|
| 0. Contract prototype | Example files, JSON schema, validator fixtures, security review | 3–5 days |
| 1. Art Lab Space Design MVP | Grid, modules, connections, chapters, placement preview, export | 8–12 days |
| 2. Inception Space importer | Upload/local import, validation report, personal-draft preview | 6–9 days |
| 3. Branching runtime | Module graph, portals, adjacent-room streaming, return route | 7–12 days |
| 4. Classroom workflow | educator review, share state, revision record, withdrawal path | 5–8 days |
| 5. Polish | map, wayfinding, mobile/tablet editing, performance and accessibility QA | 6–10 days |
An end-to-end classroom MVP is approximately 5–8 engineering weeks, plus educator review and student usability sessions. A visual contract prototype should come first; it will reveal whether Art Lab's existing canvas/selection tools can be reused before committing to the full estimate.
Parts 2 and 3 — A drafted room, its LOE, and the capture tooling research
Carried over verbatim from docs/FUTURE-PLAN.zh.md. It is two things under one old title: an effort estimate for a room that was drafted and never built, and a body of research into how children could capture motion and real objects into 3D. Neither has been acted on, nothing here has been bought, and the room has no entry in [ROOM-PLAN-TEMPLATE.md](ROOM-PLAN-TEMPLATE.md) form — writing one is the first step if it is ever picked up.
The Chinese passages are development notes; the museum runtime is English only.
Lucas Room Learning and LOE
Room Name
Technology & Biology Research Station
Summary
A giant futuristic research station is divided into two worlds. One explores advanced technology and space engineering, while the other explores biology and nature. Players can freely travel between both worlds and discover how technology and life work together.
Learning Path Categories
| Category | Fits? | Ideas |
|---|---|---|
| 🤖 AI Coding | ⭐⭐⭐⭐⭐ | AI assistants, robotics, automation, programming |
| 💻 Computer Science | ⭐⭐⭐⭐⭐ | Hardware, networking, gravity simulation |
| 🚀 Space Science | ⭐⭐⭐⭐⭐ | Spaceships, black holes, orbital mechanics |
| 🔬 Biology | ⭐⭐⭐⭐⭐ | Ecosystems, plants, evolution, biodiversity |
| 🎨 Art | ⭐⭐⭐ | Sci-fi environmental design, concept art |
| 🎮 Game Editor | ⭐⭐⭐⭐ | Build labs, quests, interactive experiments |
| 🌎 Earth Science | ⭐⭐⭐ | Environmental systems |
| 🧬 STEM | ⭐⭐⭐⭐⭐ | Cross-disciplinary science |
| 🛠 Engineering | ⭐⭐⭐⭐⭐ | Gravity generator, spacecraft systems |
| 🌱 Nature | ⭐⭐⭐⭐ | Jungle exploration |
| 🧱 3D Modeling | ⭐⭐⭐⭐ | Spaceships, research stations, jungle assets |
| 🎥 Live Painting | ⭐⭐ | Paint jungle or spaceship environments |
如果这是给 Lucas Academy / IS (Interactive Studio) 做 roadmap,我建议不要只按「功能」估算,而是按资产(Assets)+ 系统(Systems)+ 内容(Content) 来拆。这样以后所有房间都可以复用。
下面是一份比较现实的估算,假设:
- 已经有 IS 的基础框架(房间、编辑器、玩家等)
- AI Agent(Codex、Claude、GPT)负责大量编码和生成
- 人工主要负责设计、审核、测试、微调
- 单位是 工程小时 (Engineering Hours)
| Category | Task | Agent Coding | Manual | Total |
|---|---|---|---|---|
| 📐 Design | Room layout & gameplay design | 2 | 6 | 8 |
| 📐 Design | Learning path planning | 1 | 3 | 4 |
| 🏗 Environment | Main station architecture | 8 | 4 | 12 |
| 🌳 Environment | Biology jungle | 10 | 5 | 15 |
| 🚀 Environment | Technology lab | 8 | 4 | 12 |
| 🚀 Environment | Dropship hangar | 5 | 3 | 8 |
| 🎨 Art | AI concept generation | 3 | 2 | 5 |
| 🎨 Art | Asset cleanup / selection | 0 | 6 | 6 |
| 🧱 3D Modeling | Custom models | 10 | 8 | 18 |
| 🧩 Gameplay | Fork navigation | 2 | 2 | 4 |
| 🧩 Gameplay | Gravity generator interaction | 6 | 3 | 9 |
| 🧩 Gameplay | Floating transport pod | 8 | 4 | 12 |
| 🎬 Gameplay | Dropship escape sequence | 12 | 6 | 18 |
| 🎬 Gameplay | Black hole event | 10 | 5 | 15 |
| 📚 Education | Technology learning content | 4 | 8 | 12 |
| 📚 Education | Biology learning content | 4 | 8 | 12 |
| 🎵 Audio | Music & SFX integration | 2 | 4 | 6 |
| ✨ Polish | Lighting & VFX | 8 | 6 | 14 |
| 🧪 QA | Playtesting & bug fixing | 8 | 10 | 18 |
| 📖 Documentation | Docs & tutorials | 2 | 4 | 6 |
Summary
| Work Type | Hours |
|---|---|
| 🤖 Agent Coding | 113 h |
| 👨 Human Manual | 101 h |
| Total | 214 h |
如果进一步使用 AI(推荐)
很多工作其实可以进一步自动化,例如:
| Task | Manual Today | With AI Pipeline |
|---|---|---|
| Concept Images | 2 h | 0.5 h |
| Documentation | 4 h | 1 h |
| Educational Content | 16 h | 4 h |
| Environment Scripts | 24 h | 10 h |
| Gameplay Scripts | 38 h | 18 h |
| QA Test Generation | 10 h | 4 h |
这样总工时可以降到大约:
| Work Type | Hours |
|---|---|
| 🤖 Agent Coding | 130–140 h |
| 👨 Human Review / Design | 45–60 h |
| Total | 175–195 h |
我更推荐再增加一列 Owner,因为这是长期维护文档时最有价值的信息:
| Category | Task | Agent | Human | Total | Owner |
|---|---|---|---|---|---|
| Environment | Biology Jungle | 10 | 5 | 15 | AI + Artist |
| Gameplay | Dropship Escape | 12 | 6 | 18 | AI + Engineer |
| Education | Biology Lessons | 4 | 8 | 12 | Teacher |
| QA | Playtesting | 8 | 10 | 18 | Community |
| Art | Concept Images | 3 | 2 | 5 | AI |
这样的表格以后每个 Draft Room 都可以复制一份,最后还能统计整个 Lucas Academy 世界的开发预算和资源投入。
动捕
如果你的目标是给小朋友学习 3D、动画、游戏开发,其实现在已经不一定需要传统几千美元的动捕服了。
我建议按预算分成 4 档:
| 预算 | 方案 | 能做什么 | 推荐指数 |
|---|---|---|---|
| 免费-$50 | AI视频动捕(Rokoko Vision、Move AI、Wonder Studio等) | 3D动画、Blender、Unity | ⭐⭐⭐⭐⭐ |
| $300-500 | Sony Mocopi | 实时动捕、VR、Unity、Unreal | ⭐⭐⭐⭐⭐ |
| $700-1200 | HTC Vive Tracker 全身追踪 | VR游戏、动作捕捉 | ⭐⭐⭐⭐ |
| $2000+ | Rokoko、Perception Neuron | 专业动画制作 | ⭐⭐⭐ |
方案1:几乎免费(我最推荐)
现在 AI 已经改变了动捕。
孩子只需要:
- 一个手机
- 或普通 Webcam
- 穿普通衣服
就可以生成 FBX 动画。
例如:
- Rokoko Vision
- DeepMotion
- Move AI
这些都可以导入
- Blender
- Unity
- Unreal
还能直接绑定 Mixamo 人物。
Rokoko 现在提供免费的入门方案和低价 AI 动捕订阅,不需要购买硬件。([Rokoko][1])
对于 Lucas Academy 来说,这种方式特别适合:
孩子跳一下
↓
AI 自动生成角色动画
↓
放进游戏
学习体验非常棒。
方案2:Sony Mocopi(我觉得最适合教育)
这是目前消费级最成熟的方案之一。
Sony mocopi Motion Capture System
$449.99
特点:
- 6个小传感器
- 不需要摄影棚
- 不需要摄像头阵列
- 一个人几分钟就能戴好
- 可以实时驱动 Unity
- 可以驱动 Blender
- 可以做 VTuber
- 可以玩 VRChat
价格约 $450。
还有 12 传感器专业版约 $1180。
对于学校来说,一个班买一套已经够玩很多课程。
方案3:HTC Vive Tracker
如果孩子还想玩 VR:
可以买
- Meta Quest
- SteamVR
- Vive Tracker
实现全身追踪。
预算大约
- Tracker ×3
- Base Station ×2
总共约 700~1000 美元。
优点:
- VR 游戏
- VRChat
- Unity
- Unreal
缺点:
不能直接生成很干净的动画,需要后处理。
方案4:真正的专业动捕服
例如:
- Rokoko Smartsuit
- Perception Neuron
- Xsens
价格:
- Rokoko:约 $2300-$5000
- Xsens:通常 $5000-$30000+
虽然效果最好,但对于孩子学习来说性价比并不高。([Rokoko][2])
如果是 Lucas Academy,我反而会这样设计课程
Level 1(免费)
孩子:
- 手机录像
- AI 自动动捕
- Blender 导入
学习:
- 动画
- 骨骼
- 游戏角色
Level 2($450)
Sony Mocopi
孩子可以:
- 自己演
- 自己录
- 自己控制角色
做小游戏。
Level 3(多人)
几个人一起:
- 一个演骑士
- 一个演怪兽
- 一个演NPC
实时进入 Unity。
这种体验非常震撼。
我认为最值得买的是
如果预算有限,我会按下面的顺序:
- 免费 AI 动捕(Rokoko Vision 等) ⭐⭐⭐⭐⭐
- Sony Mocopi(约 $450) ⭐⭐⭐⭐⭐
- Meta Quest + Vive Tracker(如果重点是 VR)
- Rokoko 动捕服(预算充足时再考虑)
3D 建模
结合你之前规划的 Lucas Academy AI + Game + 3D 建模课程,我甚至会把动捕设计成一个完整学习路径:
真人动作 → AI 动捕 → Blender 编辑 → Unity 导入 → 游戏角色动起来
整个流程孩子通常在半小时内就能看到自己的动作出现在自己制作的游戏里,学习成就感会非常强。
[1]: https://www.rokoko.com/products/vision?utm_source=chatgpt.com "Rokoko Vision 3.0: AI Motion Capture Tool, Rebuilt" [2]: https://www.rokoko.com/products/motion-capture?utm_source=chatgpt.com "Motion capture"
如果目标是帮助孩子或初学者快速构造逼真的 3D 模型,现在最有效的方式其实已经不是传统的「一点一点建模」,而是 AI + 扫描 + 少量手工修改。
我会按效果和成本排序:
| 方法 | 成本 | 学习难度 | 真实度 | 推荐 |
|---|---|---|---|---|
| 手机照片生成3D(Photogrammetry) | 免费-$20/月 | ⭐ | ⭐⭐⭐⭐⭐ | ⭐⭐⭐⭐⭐ |
| AI 图片生成3D | 免费-$30/月 | ⭐ | ⭐⭐⭐⭐ | ⭐⭐⭐⭐⭐ |
| 深度相机扫描 | $150-500 | ⭐⭐ | ⭐⭐⭐⭐⭐ | ⭐⭐⭐⭐ |
| LiDAR 扫描 | $1000+(或部分 iPhone/iPad 自带) | ⭐⭐ | ⭐⭐⭐⭐⭐ | ⭐⭐⭐⭐ |
| 手工 Blender 建模 | 免费 | ⭐⭐⭐⭐ | ⭐⭐⭐⭐⭐ | ⭐⭐⭐ |
1. 手机拍照生成 3D(最推荐)
孩子拿手机绕着一个物体拍 30~100 张照片,例如:
- 玩具
- 花
- 树叶
- 石头
- 乐高
- 毛绒玩具
软件就能自动生成:
- Mesh
- Texture
- PBR 材质
导入:
- Blender
- Unity
- Unreal
- Roblox
- Minecraft Mod
都可以继续使用。
适合课程:
找一个校园里的东西 → 扫描 → 放进自己的游戏。
2. AI 一张图片生成 3D
现在很多 AI 已经支持:
一张图片 → 3D 模型
或者:
一句话 → 3D 模型
例如可以生成:
- 小龙
- 城堡
- 机器人
- 树
- 花
- 家具
然后导入 Blender 修改。
这种方式特别适合孩子,因为他们可以先画一张画,再把画变成立体模型。
3. 深度相机扫描
如果预算有几百美元,可以买带深度感知的设备。
例如:
- Intel RealSense(二手价格比较友好)
- Orbbec 深度相机
优点:
- 比普通照片更稳定
- 可以扫描人体
- 可以扫描家具
很多 Maker Space 都在用。
4. LiDAR(如果已经有支持的 iPhone/iPad)
如果已经有:
- iPhone Pro
- iPad Pro
就已经拥有不错的扫描设备。
可以扫描:
- 房间
- 教室
- 桌子
- 沙发
甚至整个教室都可以扫描成 3D 世界。
对于游戏课程特别有趣:
孩子扫描自己的房间,再放一个自己做的游戏角色进去。
5. AI + Blender(我认为未来的主流)
以前:
建模
↓
贴图
↓
绑定
↓
动画
未来越来越像:
AI 生成
↓
Blender 微调
↓
AI 自动贴图
↓
AI 自动绑定
↓
AI 自动动画
孩子学习的重点也会从“每个顶点怎么拉”,转向理解模型结构、比例、材质和动画设计。
如果是 Lucas Academy,我会这样设计课程
对于你规划的课程体系,我会把 3D 学习拆成几个阶段:
- 观察世界:手机扫描真实物体,理解网格(Mesh)和纹理(Texture)。
- 创造世界:AI 根据文字或图片生成角色、道具和场景。
- 编辑世界:在 Blender 中学习简单修改,例如缩放、雕刻、重新上色。
- 让它动起来:自动绑定骨骼(Rigging)、动作捕捉(Motion Capture),导入 Unity 或 Godot。
- 放进游戏:完成一个可交互的小型 3D 游戏。
这样孩子既能体验 AI 的高效率,也能逐步理解 3D 建模的基本原理,而不是一开始就被复杂的软件操作劝退。
3D人物
你现在的设备其实已经很好了:
- ✅ Wacom 数位板 → 雕刻输入
- ✅ iPhone LiDAR → 可以辅助扫描
- ✅ iPhone 相机 → 可以做 Photogrammetry
对于你的目标(人物建模练习:宝宝头、手 → 雕刻 → 游戏角色),我建议不要把 Photogrammetry 整个丢给 Codex。原因是它不是一个纯代码问题,而是一个工具链 + 艺术流程问题。Codex 可以帮你自动化部分流程,但模型生成最好用成熟软件。
我建议这样:
推荐 Pipeline(接近游戏行业)
iPhone 拍照片
|
v
Photogrammetry 软件
|
v
Raw Mesh + Texture
|
v
Blender / ZBrush 雕刻
|
v
Retopology
|
v
Unity / Unreal
1. Photogrammetry 软件推荐
🥇 RealityCapture(最推荐)
RealityCapture
现在非常强。
优点:
- 自动匹配照片非常强
- 速度快
- 人物/物体扫描效果好
- 输出质量高
流程:
导入照片
↓
Align Images
↓
Generate Mesh
↓
Texture
↓
Export OBJ/FBX
适合你:
✅ 婴儿头 ✅ 手 ✅ 玩具 ✅ 雕塑
🥈 Meshroom(免费)
Meshroom
优点:
- 免费
- 开源
- 教育价值高
缺点:
- 慢
- 对照片要求高
适合 Lucas Academy:
可以作为课程:
"AI如何从照片理解三维世界"
🥉 Polycam(最方便)
Polycam
手机 App。
优点:
- 简单
- 小孩也能用
- 直接生成模型
缺点:
- 免费版限制
- 精细控制少
适合快速体验。
2. LiDAR 怎么用?
你有 iPhone LiDAR,这是优势。
但是不要把它当主要模型来源。
更好的组合:
LiDAR负责:
大结构
例如:
- 房间
- 身体轮廓
- 桌子
- 场景
Photogrammetry负责:
细节
例如:
- 脸
- 手
- 玩具
甚至可以:
LiDAR扫描身体
+
照片扫描脸
↓
合并 Blender
这其实接近专业流程。
3. Blender 雕刻流程
导入后:
通常会得到:
500万 polygon
不要直接雕。
流程:
第一步
Decimate(降低)
↓
第二步
Voxel Remesh
↓
第三步
Sculpt
主要 Brush:
- Grab
- Clay Strips
- Smooth
- Inflate
- Crease
4. Codex 可以帮什么?
Codex 不适合:
❌ "帮我生成一个真实宝宝模型"
因为这是 AI 图形模型问题。
但是非常适合:
自动化工具
例如:
写 Blender Python:
- 自动导入扫描模型
- 自动设置材质
- 自动生成 UV
- 自动批量导出 FBX
- 自动创建 Unity prefab
例如:
你以后 Lucas Academy 可以做:
孩子上传照片
↓
后台自动:
RealityCapture
↓
Blender Python cleanup
↓
Unity角色生成
这里 Codex 很有价值。
5. 如果现在就开始,我会这样做
你的设备:
✅ iPhone ✅ Wacom ✅ Mac/PC(假设)
第一周:
Polycam
↓
扫描一个玩具头
↓
Blender Sculpt
第二周:
RealityCapture
↓
扫描真人头部
↓
修脸
第三周:
Mixamo自动绑定
↓
Unity人物
还有一个方向我觉得非常适合你的 Lucas Academy:
孩子画画 → AI生成角色 → Blender雕刻 → 动捕 → 游戏
例如:
孩子画一个“小恐龙”
↓
AI生成概念图
↓
3D生成
↓
自己用 Mocopi 动起来
↓
放进小游戏
这个完整闭环会非常有教育价值。
你现在已有的 Painterly Chameleon、游戏、AI课程,其实和这个方向非常契合。
Part 4 — [FUTURE] items in the product direction
[SPACE-MUSEUM-DIRECTION.md](SPACE-MUSEUM-DIRECTION.md) marks its own design-only sections [FUTURE], meaning recorded now and explicitly not executed. They are listed here so this file is the one place to look for unscheduled ideas:
- §6b — free web editors for models and collision volumes. An owner question from 2026-08-02, answered as a survey (Blockbench, Blender, Blender Sculpt) rather than as a decision. No tool has been adopted.
- §8 — the wider avatar pose set. The four GLB loops ship; gestures and viewing-art idle variants remain future work, and so does the blending between them.
- §9 — the "PropHunt" branch idea. Recorded only. It needs 3D modelling that does not exist, and it is not part of the museum's product line.
- §10 — multiplayer beyond presence. The presence model itself is built and documented; the direction's larger multiplayer server is a direction, and its implementation is gated.
Anything in that document not marked [FUTURE] is built — see the archive index at [archive/README.md](archive/README.md) for where each piece of it was recorded as it landed.