编程笔记

lifelong learning & practice makes perfect

Building Custom Applications on the AgiBot Lingxi X2 Robot: A Developer's Guide from Zero to One

发表于 更新于 分类于
本文字数： 3.3k 阅读时长 ≈ 3 分钟
This is the overview of the AgiBot Lingxi X2 custom development series. It introduces the on-device AIMDK, the cloud LinkSoul AgentSDK, and the learning paths for combined on-device/cloud development and Agents for complex scenarios.

A hands-on series of six articles. It covers the on-device AIMDK, the cloud LinkSoul AgentSDK, and the architecture of Agents that combine on-device and cloud capabilities in complex scenarios.

A note on terminology: AgiBot’s documentation calls this secondary development. This series uses custom development instead, because it describes what you actually build on top of the SDK. You’ll see both terms, always meaning the same thing.

What Problem This Solves

Building on the Lingxi X2 platform is not one SDK, but two parallel systems:

  • On-device AIMDK (ROS 2 Humble) — real-time control, perception, and audio capture. Code runs on the robot itself.
  • Cloud LinkSoul AgentSDK (WebSocket) — semantic understanding, LLM, and TTS. Code runs on your own server.

The two can work together or be used independently. This series advances in the order of “get it running → get it running correctly → get it running in complex scenarios”, and every article maps to a verifiable outcome.

Reading Path

1
2
3
4
5
6
7
graph LR
    A["01 Overview: the X2 platform"] --> B["02 On-Device Environment"]
    A --> D["04 Cloud AgentSDK"]
    B --> C["03 On-Device Interfaces in Practice"]
    C --> E["05 On-Device + Cloud"]
    D --> E
    E --> F["06 Agents for Complex Scenarios"]

Choose your entry point based on the resources you have:

Your Situation Path Notes
Robot only 01 → 02 → 03 The on-device chain works independently
Cloud account only 01 → 04 The full flow can be verified with text alone; no physical robot needed
Both 01 → 02 → 03 → 04 → 05 → 06 02/03 and 04 can proceed in parallel
Architecture design only 01 → 04 → 05 → 06 Skip the physical robot operations

Articles

All titles below are translated from the Chinese series; the corresponding Chinese posts are linked.

# Title Outcome Physical Robot Required
01 Understanding the Lingxi X2 and Its Development Landscape Route selection judgement + a mental model of the on-device and cloud platforms No
02 Setting Up the On-Device Environment and Your First AIMDK Program A runnable ROS 2 state query node Yes
03 On-Device Interfaces in Practice: Control, Interaction, and Perception A motion choreography script + an on-device interface cheat sheet Yes
04 Cloud AgentSDK: Passive Interaction and Semantic Skills A compliant cloud responding Agent No
05 Combining On-Device and Cloud: Voice Intent to Control Motion and Navigation A fixed joint on-device/cloud project + the ability to locate defects Yes
06 From Demo to Product: Intelligent Agents in Complex Scenarios A four-layer closed-loop architecture + two production cases No

Technical Prerequisites

Topic Required Level
ROS 2 Understanding the three concepts of node / service / topic is enough (used in article 02)
Python 3 Threads, async, and JSON processing
WebSocket Basic concepts of long connections, heartbeats, and reconnection (covered in article 04)

Global Conventions

These three appear in every article:

  1. Do not deploy custom development programs on the motion control computing unit (PC1, 10.0.1.40) — it handles highly real-time tasks such as motion control, and consuming its compute directly affects walking safety.
  2. Cross-board ROS services must include exception handling and fast retries — this is an explicit requirement in the official documentation, and extra care is needed in real scenarios.
  3. app_key / app_secret must never be hard-coded into code or written into logs — if leaked, they must be reset immediately on the LinkSoul open platform.

References

欢迎关注我的其它发布渠道