Skip to content

YIMO691/Unity-AI-LearningDemo

BranchesTags

Repository files navigation

Unity 6 AI — Combat Demo

Unity License AI Status

中文版本

A step-by-step learning repository for building a 3D AI combat demo in Unity 6 — progressing from traditional game AI (FSM, NavMesh, perception) through to modern ML inference (ML-Agents, Sentis/ONNX).

Table of Contents

Overview

This repository builds a complete AI combat demo in phases — one player versus multiple enemies. Each phase introduces a new AI technique, progressing from simple finite state machines to trained neural network inference at runtime.

Every milestone produces runnable, verifiable, and documented output. The project is AI-assisted via Claude Code + DeepSeek and remotely controllable via Feishu + OpenClaw.

AI Techniques Covered

Phase Technology What You'll See
M1 Traditional AI FSM, NavMesh, Perception Enemies patrol, detect, chase, and attack
M2 Behavior & Swarm Unity Behavior, Group Behavior Coordinated squad tactics, dynamic decision trees
M3 ML-Agents PPO, TensorBoard Self-trained agents with observable reward curves
M4 Sentis Inference ONNX, Burst CPU Real-time neural network inference in-game
M5 Integration & Delivery CI, Demo Video Reproducible full demo from scratch

Detailed roadmap: docs/planning/learning-roadmap.md

Current Status

M0, M1, and M2 are complete. M3 (ML-Agents Reinforcement Learning) is ready to begin.

M0 — Project Foundation ✅

Item Status
Unity 6 URP Project ✅
Package Installation ✅ 5 core packages
Base Scene + Smoke Test ✅
Git LFS + CI + Docs ✅

M1 — Traditional AI (FSM) ✅

Feature Status
Patrol / Chase / Attack / Death ✅
Raycast occlusion ✅
Scene Demo1_FSM.unity

M2 — Behavior Tree + Swarm ✅

Feature Status
C# Behavior Tree (Selector/Sequence) ✅
Swarm separation (Boids 3 rules) ✅
Unity Behavior evaluated → C# BT chosen ✅
Scenes Demo2_Behavior.unity / Demo2_Swarm.unity

Full status: PROGRESS.md | TODO.md

Milestones

Milestone Status Deliverables
M0 Project Foundation ✅ Complete Unity 6 skeleton, packages, CI, docs system
M1 Traditional Game AI ✅ Complete FSM + NavMesh patrol/chase/attack, Demo1_FSM scene
M2 Behavior Tree & Swarm ✅ Complete C# BT + Boids separation, Demo2 scenes
M3 ML-Agents Training 🟡 In Progress PPO training, TensorBoard, ONNX model
M4 Sentis Inference ⏳ Pending ONNX import, Burst CPU runtime inference
M5 Final Delivery ⏳ Pending Full demo, CI green, video walkthrough

Next Steps / 下一步

M3 — ML-Agents Reinforcement Learning

Build a training environment where an agent learns combat behavior through PPO (Proximal Policy Optimization).

Task Description
Python Setup Install mlagents Python package for training
Training Scene Create TrainEnv.unity with Agent + environment
Custom Agent Define observations (player position, distance), actions (move, attack), rewards (hit=+1, miss=-0.1)
PPO Training Write YAML config, run mlagents-learn, monitor TensorBoard
ONNX Export Export trained model, verify inference in Unity

Prerequisite: Python 3.10+ with mlagents package installed.

Detailed plan: M0 Task Plan | Architecture Decisions

Quick Start

Prerequisites

Tool Version Purpose
Unity Hub + Unity 6 6000.4.5f1+ Editor & runtime
Git 2.40+ Version control
Git LFS 3.x Large binary storage
GitHub CLI (gh) 2.x PR, Issue, CI management
PowerShell 5.1+ Automation scripts

Clone & Setup

git clone https://github.com/YIMO691/Unity-AI-LearningDemo.git
cd Unity_AI
git lfs install

Open in Unity

  1. Launch Unity Hub
  2. Click Open → select Unity_AI/UnityProject/
  3. Wait for package restoration and script compilation
  4. Open Assets/_Project/Scenes/Main.unity
  5. Press Play — should see gray ground with lighting, no Console errors

Run Tests

Window → General → Test Runner → EditMode → Run All

Should show SmokeTest.Project_CanRunEditModeTest ✅ 1/1 passing.

Project Status Check

powershell -ExecutionPolicy Bypass -File ".\tools\mobile-status.ps1"

AI Assistant Setup

$env:DEEPSEEK_API_KEY="sk-your-key"
.\scripts\start-claude-deepseek.cmd

⚠️ Never commit API keys. See .env.example for configuration reference.

Project Structure

F:\Unity6_AI
├─ .claude/                  # Claude Code skills, rules, commands
├─ .github/                  # CI workflows, PR/Issue templates, CODEOWNERS
├─ .openclaw/                # OpenClaw Gateway state & agent config
├─ docs/                     # All project documentation
│  ├─ planning/              # Plans, milestones, ADR, package manifest
│  ├─ status/                # PROGRESS.md, TODO.md, AI_DEV_LOG.md
│  ├─ reference/             # Manuals, conventions, skills manifest
│  ├─ workflows/             # Setup guides (CI, GitHub, Feishu, Claude Code)
│  ├─ environment/           # Environment reports
│  └─ templates/             # Reusable file templates
├─ LearningBlog/             # Structured learning notes
├─ Logs/                     # Runtime logs (not committed)
├─ prompts/                  # AI session starter prompts
├─ scripts/                  # Claude Code + DeepSeek startup scripts
├─ tools/                    # Automation PowerShell scripts
│  ├─ claude-code-relay.ps1  # Relay: Feishu → Claude Code
│  ├─ mobile-status.ps1      # Phone-friendly project status
│  └─ cc-*.ps1              # /cc, /cc-run, /cc-status, /cc-session commands
├─ UnityProject/             # Unity 6 project root
│  └─ Assets/_Project/       # Project source code & assets
│     ├─ Art/                # Materials, models, textures
│     ├─ Prefabs/            # Reusable prefabs
│     ├─ Scenes/             # Main.unity and future scenes
│     ├─ Scripts/            # C# scripts (Core/AI/Player/Training)
│     └─ Tests/              # EditMode + PlayMode tests
├─ LICENSE                   # MIT license
├─ CLAUDE.md                 # Claude Code project instructions
├─ AGENTS.md                 # AI agent handoff notes
├─ CONTRIBUTING.md           # Contribution guide
├─ SECURITY.md               # Security policy
└─ README.md                 # This file

Remote Development

Control this project from your phone via Feishu + OpenClaw. Check progress, dispatch coding tasks, and monitor CI — all through chat. The relay architecture enforces safety boundaries: read-only commands never modify files, write commands require explicit authorization.

Feishu Mobile → OpenClaw Gateway → unity6ai Agent → Claude Code Relay → Claude Code + DeepSeek → Result → Feishu
Command Mode Function
/cc Read-only Project status, TODO check, log analysis
/cc-run <task> Edit (restricted) Execute task via relay with safety rules
/cc-run-big <task> Edit (unrestricted) Direct Claude Code execution for large tasks
/cc-status Read-only Current Claude Code task state
/cc-last Read-only Last execution summary
/progress Read-only Mobile-optimized project progress

Setup guide: Feishu + OpenClaw Deployment Relay guide: Claude Code Relay

Engineering Foundations

Foundation File Purpose
Architecture Decisions ADR 8 technical decisions with rationale
File Organization Convention Where every file type belongs
Skills Manifest Manifest 20 active skills with decision flow
Coding Rules C# Rules Unity C# conventions, Inspector, validation
Git LFS .gitattributes Binary file tracking (ONNX, FBX, textures, audio)
CI Pipeline unity-ci.yml GitHub Actions: compile + EditMode tests
Package Versions Manifest Locked versions for reproducibility
Security Policy SECURITY.md Vulnerability reporting & scope

Documentation

Entry Points

Document Purpose
CLAUDE.md Claude Code project instructions
AGENTS.md AI agent handoff notes
CONTRIBUTING.md Contribution guide & PR checklist
SECURITY.md Security policy & reporting
LICENSE MIT license

Planning

Document Purpose
project-brief.md Project requirements & milestone summary
learning-roadmap.md Phased learning path
milestone-m0-plan.md M0 detailed task breakdown
architecture-decisions.md Architecture Decision Records (8 ADRs)
package-manifest.md Locked Unity package versions

Status Tracking

Document Purpose
PROGRESS.md Current progress & completed items
TODO.md Task backlog & acceptance criteria
AI_DEV_LOG.md AI collaboration log

Reference

Document Purpose
manual.md Original learning manual (read-only archive)
file-organization-convention.md File placement rules & root whitelist
skills-manifest.md 20 skills inventory with decision flow

Workflows

Document Purpose
claude-deepseek-workflow.md Claude Code + DeepSeek setup
github-setup.md GitHub repository creation
feishu-openclaw-deployment.md Remote development setup
claude-code-relay.md Claude Code relay script guide
skills-strategy.md Skills/agents/hooks adoption strategy
docs/README.md Documentation index & audit

Learning Blog

Document Topic
LearningBlog/README.md Blog index by date & tag
Toolchain Setup SSH Key, Git LFS, GitHub CLI deep-dive
Remote Dev Chain OpenClaw + Feishu + Claude Code relay

Contributing

This is a personal learning project. Suggestions and discussions are welcome via GitHub Issues. Code contributions should follow the conventions in CONTRIBUTING.md, CLAUDE.md, and .claude/rules/unity-csharp.md.

Branch Naming

milestone/m{N}-{slug}
# Examples: milestone/m1-fsm, milestone/m2-behavior

Code Conventions

  • Communication in Chinese, identifiers in English
  • [SerializeField] private over public fields
  • AI parameters exposed in Inspector
  • After writing code, always explain: which GameObject, which components, how to verify
  • Never commit: API keys, Unity Library, Temp, training artifacts, user settings

License

MIT — see LICENSE for details.

Built with Unity 6 | AI-assisted by Claude Code + DeepSeek | Remote-controlled via Feishu + OpenClaw

About

No description, website, or topics provided.

Resources

Readme

License

MIT license

Contributing

Contributing

Security policy

Security policy
Activity

Stars

0 stars

Watchers

0 watching

Forks

0 forks
Report repository

Releases

No releases published

Packages

 
 
 

Contributors

Languages