跳到主要内容

Pilot to Production Deployment

Overview​

Atlas Pilot to Production Deployment 是 OEM 或 Sensor Manufacturer 将 Atlas 从受控试点逐步扩展为长期 Runtime Governance Infrastructure 的项目管理框架。

它不是一次性软件安装，也不是围绕单一机器人、单一 Sensor 编写的内部 Glue Code。

Atlas 的目标是建立一套可跨：

  • Robot SKU
  • Sensor SKU
  • SBC / Host Platform
  • Linux / ROS Environment
  • Deployment Scenario
  • OEM 与 Sensor Manufacturer

持续复用的 Runtime Governance Infrastructure。

Pilot
  │
  ▼
Controlled Deployment
  │
  ▼
Production Readiness
  │
  ▼
Multi-SKU Expansion
  │
  ▼
Full-scale Deployment

每个阶段都必须具备独立 Scope、Budget、Milestone、Acceptance Criteria 和退出机制。

Pilot 成功不代表自动进入全量部署。

Infrastructure, Not Glue Code

Atlas 不应被部署为只适用于某一款机器人的临时代码。

典型 Glue Code 模式是：

Robot A
+
Sensor X
+
One Driver
+
One Script
+
One Customer Issue

当 Robot、Sensor、SBC 或 Software Version 改变时，工程团队需要重新开发和维护。

Atlas 采用基础设施模式：

Canonical Runtime Surface
+
Standard Runtime Dataset
+
Standard Evidence Pack
+
Standard Investigation Workflow
+
Reusable Historical RGA

具体产品差异通过以下方式接入：

  • Surface Registry
  • Runtime Adapter
  • Configuration
  • Product Profile
  • Deployment Policy

Atlas Core、Evidence Schema、Investigation Chain 和 RGA Model 保持一致。

Cross-SKU and Cross-platform Architecture

Atlas 应支持在统一治理模型下扩展到不同平台。

Atlas Runtime Governance Infrastructure
│
├── Robot SKU A
│   ├── SBC A
│   ├── LiDAR X
│   └── Camera Y
│
├── Robot SKU B
│   ├── SBC B
│   ├── LiDAR Z
│   └── IMU M
│
└── Robot SKU C
    ├── Industrial PC
    ├── Camera Y
    └── Radar N

跨平台不表示所有平台具有完全相同的 Runtime Surface。

它表示：

  • 使用统一的 Surface Definition
  • 使用统一的 Evidence Pack Schema
  • 使用统一的 Investigation Objects
  • 使用统一的 Historical RGA Model
  • 对未覆盖 Surface 明确标记
  • 通过 Adapter 处理接口差异

Dual-flywheel Deployment Model

Atlas 采用 OEM 与 Sensor Manufacturer 双飞轮设计。

OEM Runtime Governance Flywheel
            │
            │ Sensor Engagement
            ▼
Sensor Runtime Governance Flywheel
            │
            │ Sensor IR / LL / RGA
            ▼
OEM Runtime Governance Flywheel

无论最先采用 Atlas 的是 OEM 还是 Sensor Manufacturer，Atlas 都可以沿着真实调查协作链继续扩展。

最终目标是逐步建立：

Sensor
   ↓
Bus / Power / Timing
   ↓
Linux / Driver
   ↓
SBC
   ↓
ROS / Application
   ↓
Robot Runtime

这一 Sensor-to-SBC Runtime Governance Ecosystem。

OEM-first Expansion​

当 OEM 首先采用 Atlas：

OEM Robot Deployment
        │
        ▼
OEM REF Investigation
        │
        ▼
Sensor Candidate
        │
        ▼
Sensor Engagement Pack
        │
        ▼
Sensor FAE Investigation
        │
        ▼
Sensor Historical RGA

Atlas 可以继续协助相关 Sensor Manufacturer 建立：

  • Sensor Runtime Profile
  • Sensor Historical RGA
  • Standard FAE Investigation Workflow
  • OEM Collaboration Interface

Sensor-first Expansion​

当 Sensor Manufacturer 首先采用 Atlas：

Sensor Runtime Profile
        │
        ▼
Sensor Historical RGA
        │
        ▼
OEM Integration
        │
        ▼
OEM Runtime Surface Mapping
        │
        ▼
OEM REF Investigation
        │
        ▼
OEM Historical RGA

Atlas 可以沿着 Sensor 的实际 OEM 客户与机器人平台，逐步建立完整的 Sensor-to-SBC 治理链。

OEM and Sensor Deployment Differences

OEM 与 Sensor Manufacturer 使用相同 Atlas Canonical Model，但项目目标不同。

ItemOEM DeploymentSensor Manufacturer Deployment
Primary ObjectRobot SKU / FleetSensor Product / Sensor SKU
Main InvestigationRobot Runtime REFSensor REF / FAE Investigation
Surface ScopeSensor、Power、Bus、Linux、Driver、ROS、Robot StateSensor、Firmware、Interface、Driver、Output、Timing
Historical RGAMulti-sensor and system casesSensor-only cases across OEMs and environments
Main UserTier 1 / Tier 2 / Tier 3FAE / Driver / Firmware / Product Team
Expansion UnitOne Robot Model at a timeOne Sensor Product at a time
ROI FocusInvestigation time、Tier 3 involvement、fleet support costFAE response、cross-OEM reuse、engineering escalation
CollaborationEngage Sensor FactorySupport multiple OEMs

Project Governance

每一个 Pilot、Controlled Deployment 和 Production Deployment 都应作为正式项目管理。

最低项目结构：

Executive Sponsor
        │
        ▼
Organizational Representative
        │
        ▼
Project Manager
        │
        ├── Business Owner
        ├── Runtime Engineering Lead
        ├── Investigation Lead
        ├── IT / Security
        ├── Legal / Procurement
        └── Atlas Project Lead

小型项目中，一个人可以承担多个角色。

但每项责任必须有明确 Owner。

Organizational Representative

客户必须指定一名 Organizational Representative。

该角色不是普通技术联系人，而是客户组织的正式项目代表。

主要职责：

  • 确认项目范围
  • 协调跨部门资源
  • 确认客户输入
  • 批准 Change Request
  • 签署 Milestone Acceptance
  • 管理项目升级
  • 提交继续、暂停或终止决定

Atlas 不应以单一工程师的非正式意见作为组织级验收结论。

OEM Project Roster

OEM 项目建议包括：

RoleMain Responsibility
Executive SponsorBudget 与部署决策
Organizational Representative跨部门协调与正式验收
OEM Project ManagerSchedule、Risk、Milestone
Product OwnerRobot SKU 与业务场景
Runtime Engineering LeadAgent 与 Surface 接入
Tier 1 RepresentativeREF Intake
Tier 2 Investigation LeadEvidence、Recall 与调查组织
Tier 3 Engineering Representative复杂事件与技术边界
Fleet / Cloud Owner数据传输与总部接入
Sensor Coordination OwnerSensor Factory 协作
IT / Security数据、权限与部署审查
Legal / Procurement合同、IP 与退出机制

Tier 3 不应承担 Pilot 的常规日志收集和数据整理。

Sensor Manufacturer Project Roster

Sensor Manufacturer 项目建议包括：

RoleMain Responsibility
Executive SponsorBudget 与产品线决策
Organizational Representative正式协调与验收
Project ManagerSchedule、Scope、Risk
Sensor Product OwnerSensor SKU 与目标市场
FAE LeadSensor REF Workflow
Driver LeadDriver Runtime Boundary
Firmware LeadFirmware Investigation
Validation / Reliability LeadKnown Environment Coverage
OEM Collaboration OwnerOEM 接入和 EGP
IT / SecurityRepository 与数据交换
Legal / ProcurementNDA、IP 与退出机制

Budget Structure

项目预算应覆盖完整部署成本，而不仅是 Atlas License。

Total Project Budget
│
├── Atlas Software / License
├── Professional Services
├── Integration and Adapter Work
├── Historical RGA Pre-build
├── Infrastructure
├── Customer Internal Engineering
├── Training and Change Management
├── Security / Legal Review
└── Contingency

OEM Budget Items​

OEM 通常需要预算：

  • Robot Runtime Surface Mapping
  • Atlas Agent Integration
  • Fleet / Cloud Integration
  • OEM Historical RGA Pre-build
  • Tier 1 / Tier 2 Workflow Setup
  • Sensor Factory Collaboration
  • Selected Robot Deployment
  • Governance Metrics and ROI Tracking
  • Training and Operational Handover

Sensor Manufacturer Budget Items​

Sensor Manufacturer 通常需要预算：

  • Sensor Runtime Surface Mapping
  • Sensor Runtime Profile Creation
  • Known Environment / CE Observation
  • Sensor Historical RGA Pre-build
  • FAE Workflow Integration
  • Driver / Firmware Escalation Workflow
  • OEM EGP Collaboration
  • Product and Environment Expansion
  • FAE ROI Tracking

Contingency Budget​

建议预留：

10%–20%

用于已批准的：

  • Additional Adapter
  • New Surface
  • Extra Test Cycle
  • Infrastructure Change
  • Approved Scope Expansion

Contingency 不应被视为默认可消费预算。

Milestone-based Payment

建议采用 Milestone Payment。

每一期付款应绑定：

  • Defined Deliverable
  • Acceptance Criteria
  • Acceptance Period
  • Written Acceptance
  • Remediation Process
  • Change Boundary

Suggested Payment Structure​

MilestoneSuggested Payment
Contract and Project Initiation15%
Architecture and Surface Readiness15%
Historical Asset Initialization15%
Pilot Technical Chain Acceptance20%
Controlled Deployment Acceptance15%
Production Readiness Acceptance10%
Final Handover / Full-scale Plan10%

实际比例可以根据项目规模调整。

Stage 0 — Project Initiation

Common Deliverables​

  • Contract / SOW
  • Project Charter
  • Approved Budget
  • Named Roster
  • Scope Definition
  • Data and IP Boundary
  • Milestone Plan
  • Payment Schedule
  • Change Process
  • Exit Mechanism
  • Escrow Requirement

Start Gate​

项目启动前必须确认：

  • Executive Sponsor
  • Organizational Representative
  • Project Manager
  • Budget Approval
  • In-scope Product
  • In-scope Environment
  • Acceptance Signatory
  • Data Owner
  • Exit Terms

Stage 1 — Pilot Deployment

Pilot 应限制在明确的最小范围内。

OEM Pilot Scope​

建议选择：

One Robot Model
+
One Critical Sensor
+
One Deployment Scenario
+
One to Three REF Types
+
Five to Ten OEM Historical Cases

例如：

Delivery Robot A
+
LiDAR X
+
Restaurant
+
Unexpected Stop

Sensor Manufacturer Pilot Scope​

建议选择：

One Sensor Product
+
One Firmware / Driver Combination
+
Two to Four Environments
+
Three to Five Runtime Patterns
+
Five to Ten Historical Cases

例如：

LiDAR X
+
Firmware 2.4
+
Warehouse / Factory / Restaurant
+
Packet Loss / Reconnect / Timestamp Gap

Pilot Acceptance​

Pilot 应验证：

  • Runtime Surface 可观察
  • Agent 可以 Observe、Persist、Retain、Export
  • Evidence Pack 可生成
  • Historical RGA 可召回
  • Investigation Context 可形成
  • Investigation 可以闭环
  • IR 与 LL 可以保存
  • 新 RGA 可以建立
  • ROI Baseline 可以计算

Stage 2 — Controlled Deployment

Controlled Deployment 是 Pilot 与 Production 之间的正式阶段。

它不是无边界扩大。

核心原则：

每次只扩展一个主要产品模型或一个明确的平台组合。

OEM Controlled Deployment​

OEM 应采用：

One Robot Model at a Time

每一款 Robot Model 应分别确认：

  • SBC / Host Platform
  • Linux / ROS Version
  • Sensor Combination
  • Power Architecture
  • Communication Surface
  • Deployment Scenario
  • Historical RGA Coverage
  • Export Infrastructure
  • Support Roster

推荐顺序：

Robot Model A
      │
      ▼
Acceptance
      │
      ▼
Robot Model B
      │
      ▼
Acceptance
      │
      ▼
Robot Model C

不要一次将 Atlas 部署到所有 Fleet SKU。

Sensor Controlled Deployment​

Sensor Manufacturer 应采用：

One Sensor Product at a Time

每款 Sensor 应分别确认：

  • Hardware Revision
  • Firmware Version
  • Driver / SDK
  • Interface
  • Host Platform
  • Known Environment
  • Runtime Profile
  • Historical RGA
  • FAE Workflow

推荐顺序：

Sensor Product X
      │
      ▼
Acceptance
      │
      ▼
Sensor Product Y
      │
      ▼
Acceptance

同一 Sensor Product 可以逐步扩展到不同 OEM 和场景。

Controlled Deployment Gates

每次扩展前必须通过以下 Gate：

Surface Readiness

Historical RGA Readiness

Agent Stability

Evidence Chain

Investigation Workflow

Security Review

Operational Owner

ROI Tracking

未通过 Gate 的产品模型不进入下一阶段。

Stage 3 — Production Readiness

Production Readiness 评估 Atlas 是否具备长期运行条件。

评估包括：

  • Runtime Stability
  • Storage and Retention
  • Dataset Lock
  • Export Reliability
  • Repository Backup
  • Access Control
  • Historical Recall Quality
  • Ticket Closure
  • Sensor Collaboration
  • Operational Support
  • Auditability
  • Exit Readiness

Production Readiness 不代表 Atlas 对所有未来 REF 提供自动根因结论。

Stage 4 — Multi-SKU and Cross-platform Expansion

完成第一款产品的 Controlled Deployment 后，可以复用 Atlas Canonical Infrastructure。

Validated Atlas Core
        │
        ├── New Surface Mapping
        ├── New Adapter Configuration
        ├── New Product Profile
        └── New Historical RGA

不应为每个新 SKU 重写独立调查系统。

Reusable Components​

跨 SKU 复用的部分包括：

  • Atlas Agent Core
  • Runtime Dataset Lifecycle
  • Evidence Pack Schema
  • Five-Segment Window
  • REF Lifecycle
  • Historical Recall
  • Investigation Context
  • Investigation Tier Candidate
  • IR / LL Model
  • Ticket Closure
  • Governance Metrics

Product-specific Components​

每个 SKU 需要确认：

  • Surface Registry
  • Adapter
  • Product Metadata
  • Timestamp Source
  • Runtime Profile
  • Historical RGA Coverage
  • Retention Policy
  • Deployment Configuration

Stage 5 — Full-scale Deployment

Full-scale Deployment 应在多个 Controlled Deployment 成功后启动。

Validated Product Models
        │
        ▼
Standard Deployment Template
        │
        ▼
Fleet / Product-line Rollout
        │
        ▼
Central Governance Operation

OEM Full-scale Deployment​

可以包括：

  • Multiple Robot Models
  • Multiple Fleet Regions
  • Multiple Deployment Scenarios
  • Multiple Sensor Manufacturers
  • Central Tier 2 Investigation
  • Distributed Tier 1 Intake
  • Controlled Tier 3 Escalation
  • CTO Runtime Governance Dashboard

Sensor Manufacturer Full-scale Deployment​

可以包括：

  • Multiple Sensor Products
  • Multiple Firmware Branches
  • Multiple OEM Customers
  • Multiple Robot Platforms
  • Multiple Deployment Environments
  • Central Sensor RGA Repository
  • Standardized FAE Investigation
  • Product Reliability Feedback

Dual-flywheel Operating Model

Full-scale Deployment 后，两个飞轮持续互相增强。

OEM Flywheel​

OEM REF
   ↓
Evidence Pack
   ↓
Historical Recall
   ↓
Investigation
   ↓
IR / LL
   ↓
OEM Historical RGA
   ↓
Faster Future Investigation

Sensor Flywheel​

Sensor REF
   ↓
Sensor Runtime Profile
   ↓
Sensor Historical Recall
   ↓
FAE Investigation
   ↓
Sensor IR / LL
   ↓
Sensor Historical RGA
   ↓
Faster Cross-OEM Support

Flywheel Connection​

OEM Historical Knowledge
        │
        ▼
Sensor Engagement Pack
        │
        ▼
Sensor Historical Knowledge
        │
        ▼
Sensor Response
        │
        ▼
OEM Closure and Reuse

双方不需要共享完整私有 Repository。

它们通过标准对象和授权边界协同。

Change Management

以下事项必须进入正式 Change Request：

  • New Robot Model
  • New Sensor Product
  • New SBC Platform
  • New Deployment Environment
  • New Runtime Surface
  • New REF Type
  • New Cloud Integration
  • New Data Mode
  • New Geographic Region
  • New Production SLA

Change Request 应记录：

  • Scope Impact
  • Technical Impact
  • Budget Impact
  • Schedule Impact
  • Security Impact
  • Acceptance Impact

Data and IP Ownership

Customer-owned​

  • Customer Runtime Dataset
  • Raw Logs
  • Customer Configuration
  • Customer Historical Cases
  • Customer IR / LL
  • OEM Historical RGA
  • Sensor Historical RGA
  • Customer-specific Governance Records

Atlas-owned​

  • Atlas Core Code
  • Canonical Schema
  • Generic Runtime Governance Logic
  • Evidence Pack Model
  • Investigation Workflow
  • Generic Adapter Framework
  • Pre-existing Atlas IP

Project-specific Deliverables​

以下内容的所有权或使用权必须在合同中单独定义：

  • Customer-specific Adapter
  • Customer-specific Integration
  • Deployment Configuration
  • Custom Reporting
  • Custom Workflow Extension

Safe Exit Mechanism

每个阶段都必须允许客户在可控条件下退出。

Exit Decision
      │
      ▼
Freeze Accepted Project State
      │
      ▼
Export Customer Assets
      │
      ▼
Technical Handover
      │
      ▼
Revoke Access
      │
      ▼
Confirm Data Deletion

Exit Package

建议包括：

  • Runtime Dataset Export
  • Evidence Pack Export
  • Historical RGA Export
  • Sensor Runtime Profile Export
  • Surface Registry
  • Configuration
  • Investigation Records
  • Schema Documentation
  • Deployment Documentation
  • Accepted Release Information
  • Open Issue List
  • Data Deletion Confirmation

建议采用开放格式：

  • JSON
  • Markdown
  • CSV
  • Documented Archive

客户资产不应只能通过 Atlas UI 读取。

Code Escrow

Code Escrow 用于保障长期生产部署的供应商连续性。

Pilot 阶段可以先确认 Escrow 条款。

进入 Production 前完成正式 Escrow。

Escrow Scope​

可以包括：

  • Contracted Atlas Release
  • Customer-specific Adapter
  • Build Instructions
  • Dependency Manifest
  • Deployment Documentation
  • Configuration Schema
  • Verification Hash

Escrow Verification​

应验证：

  • Archive Completeness
  • Version Match
  • Hash Integrity
  • Build Reproducibility
  • Dependency Availability
  • Deployment Instructions

仅保存一个无法构建的代码压缩包不构成有效 Escrow。

Escrow Release Triggers​

可以包括：

  • Atlas Insolvency
  • Permanent Product Discontinuation
  • Contracted Critical Support Termination
  • Material Breach Not Cured
  • Agreed Business Continuity Event

普通项目争议不应自动触发 Escrow Release。

Termination and Payment

退出责任应根据原因处理。

Customer Convenience Termination​

通常支付：

  • 已验收 Milestone
  • 已完成但尚处于验收期的可交付成果
  • 已批准且不可取消的成本
  • 合同约定的 Transition Cost

未启动的未来 Milestone 不应自动全额收费。

Atlas Material Breach​

合同可以约定：

  • 暂停受影响 Milestone 付款
  • 补救期
  • 未验收款项退款
  • 强制 Exit Package
  • Transition Assistance
  • 符合条件时触发 Escrow

Controlled Deployment Exit​

某款 Robot Model 或 Sensor Product 未通过验收时，可以只停止该模型扩展。

不必自动终止已验收的其它部署。

Robot A — Accepted

Robot B — Failed Gate

Robot C — Not Started

这种模型级退出机制可以隔离扩展风险。

Management Decision Gates

每个阶段结束后，应进行正式管理层决策。

可选结果：

Approve Next Stage

Approve with Conditions

Extend Current Stage

Request Remediation

Pause Deployment

Exit Selected Model

Terminate Program

决策依据包括：

  • Technical Acceptance
  • Budget Consumption
  • Schedule
  • Security
  • Historical RGA Readiness
  • Investigation ROI
  • Tier 3 Involvement
  • Operational Readiness
  • Remaining Risk

ROI Tracking

项目应从 Pilot 开始建立 Baseline。

建议指标：

  • Time to Evidence Pack
  • Median Investigation Time
  • Ticket Closure Time
  • Historical RGA Reuse Rate
  • Tier 3 Involvement Rate
  • Engineering Hours per REF
  • Repeat Investigation Avoided
  • Sensor FAE Response Time
  • Cross-OEM RGA Reuse
  • Cost per Closed REF

OEM ROI​

重点观察：

  • Robot Model REF Trend
  • Fleet Investigation Cost
  • Tier 3 Escalation
  • Sensor Collaboration Time
  • Historical RGA Reuse
  • Customer Support Efficiency

Sensor Manufacturer ROI​

重点观察：

  • FAE Investigation Time
  • Cases per Sensor Product
  • Cross-OEM Reuse
  • Driver / Firmware Escalation
  • Repeated Question Reduction
  • Environment Pattern Coverage

Minimum Production Gate

进入 Full-scale Deployment 前，至少需要：

  • One or More Controlled Models Accepted
  • Stable Atlas Agent
  • Defined Surface Coverage
  • Historical RGA Repository
  • Operational Roster
  • Security Approval
  • Backup and Restore
  • Data Export Capability
  • Exit Package Tested
  • Code Escrow Completed, if contracted
  • ROI Baseline Established
  • Production Budget Approved

Summary

Atlas Pilot to Production Deployment 应遵循：

Pilot

↓

One-model Controlled Deployment

↓

Production Readiness

↓

Cross-SKU / Cross-platform Expansion

↓

Full-scale Deployment

OEM 以 Robot Model 为扩展单位，建立从 Sensor、Power、Bus、Linux、Driver 到 ROS 和 Robot Runtime 的完整治理链。

Sensor Manufacturer 以 Sensor Product 为扩展单位，建立 Sensor Runtime Profile、Historical RGA 和标准 FAE Investigation Workflow，并逐步扩展到不同 OEM、机器人平台和部署场景。

Atlas 的双飞轮意味着：

  • OEM 先采用，可以沿 Sensor Engagement 链引入 Sensor Manufacturer。
  • Sensor Manufacturer 先采用，可以沿 OEM Integration 链引入 Robot Runtime Governance。
  • 双方最终共同建立 Sensor-to-SBC Runtime Governance Ecosystem。

下一步阅读

  • SDK
  • ROS2
  • API
  • CLI Atlas 不是单一项目的 Glue Code。

它是一套可以跨 SKU、跨平台、跨场景和跨组织持续复用的 Runtime Governance Infrastructure。