AelosEmbodied

English | 简体中文

Aelos Embodied humanoid-robot development repository, currently at the initial stage with the vendor documentation, reference projects, and action assets as the starting point.

Aelos Embodied (乐聚 Aelos EM) is a biped humanoid robot for embodied-AI education and research. Programs are authored in Leju's aelos_edu desktop software, a block/Python hybrid IDE; the robot runs a ROS-based controller, and motion routines are stored and downloaded separately as action files. This is the development repository for the platform.

Current status: initial setup. No original development programs are included yet. The repository holds the four reference IDE projects, the vendor course material and the action packs that the work starts from; new programs will be added here as development progresses.

The included projects are IDE output, not a standalone SDK: .js, .lua and main.py are generated by aelos_edu 2.1.5+, and the Blockly workspace in test.xml is the editable source of each project. The .abe file holds project metadata only.

Repository Structure

Path	Contents
code/EM_demo/	AR-tag visual-servo box handling — two waypoints, chest-camera alignment and 20 imported carry actions
code/EM_default/	Leju factory built-in project (Chinese, 2025-03-22 image) — gamepad-selectable performance showreel
code/EM_serialcheck/	Serial-link check scaffold with a single Hands_up action block
code/try/	Empty project skeleton written by aelos_edu 2.2.8
doc/	User manual V1.0.1, three course decks (vision, AR-tag, sensors) and a printable AR-tag sheet
src/aelos_舞蹈动作文件/	Action and audio library — 29 fable and poem routines, 34 music-dance routines and 13 newer dance scripts

Reference Projects

The four aelos_edu projects are kept as references rather than as original development work.

EM_demo — AR-tag guided box handling

The most complete reference project. The chest camera reports AR-tag pose, and a bang-bang loop corrects yaw first (±10°), then the lateral offset (±0.03 m), and finally advances until the tag is within 0.07 m, steps in and places the box. Two procedures named 地点1 and 地点2 (waypoint 1 and waypoint 2) align to tags 11 and 12, separated by two large left turns while carrying. main.py records the control loop; condensed, one waypoint looks like this:

if artag_port.get_specifies_tag(11, 'chest')[0] != 1000:
    tag_x   = artag_port.get_specifies_tag(11, 'chest')[0]
    tag_y   = artag_port.get_specifies_tag(11, 'chest')[1]
    tag_yaw = artag_port.get_specifies_tag(11, 'chest')[2] + 90
    if tag_yaw > 10:      base_action.action('抱块小左转')
    elif tag_yaw < -10:   base_action.action('抱块小右转')
    elif tag_y > 0.03:    base_action.action('抱块小左侧移')
    elif tag_y < -0.03:   base_action.action('抱块小右侧移')
    elif tag_x > 0.07:    base_action.action('抱块向前慢走半步2')
    else:                 base_action.action('抱块前进一步2'); base_action.action('放下方块'); break

Twenty action modules are imported under import/ — lift, place, carry steps, side steps, turns and 翻墙1 — and src/ keeps the two hand-tuned scripts used for the final approach.

EM_default — factory showreel

The built-in Chinese project shipped with the 2025-03-22 image. A while(true) loop reads the gamepad and maps four buttons to a music track plus a short performance (鞠躬, 俯卧撑, 展翅, 左脚射门) and two more to imported full dances (少年, 科目三). Seven imported routines are embedded in the project, and the Lua action library is committed as readable text. The blocks target the on-robot Lua runtime only; no ROS/Python export is included.

EM_serialcheck and try

EM_serialcheck is the vendor's serial-link check (串口连接检查): one Hands_up (举手) action module survives, without a workspace file, so it is a scaffold for verifying that the IDE can talk to the robot. try is an empty 2.2.8 project with a single, unconnected start block; its lib.lua is the human-readable runtime library the IDE emits for a new project.

Motion and Audio Assets

src/aelos_舞蹈动作文件/ is a vendor/community pack rather than authoring code. The 29 fable and poem routines (丑小鸭, 小红帽, 江雪, 登鹳雀楼, … in Chinese and English versions) and the 34 music-dance scripts (小苹果, 江南style, 最炫民族风, let it go, ppap, …; the pack is labeled 32 and carries two variants) ship as .src scripts with matching .mp3 soundtracks and one-page list PDFs. 新版舞蹈/ adds 13 newer scripts, including 科目三, APT, 你要跳舞吗 and 勇气大爆发 (which also keeps a block-program screenshot).

A .src file is plain text — rigidity and speed lines, MOTORA–MOTORE joint angles, WAIT and DELAY. The IDE imports one as a custom action, and the Python target exposes it as base_action.action('<name>'); total assets are 76 scripts and 72 audio tracks. No open license is stated for the packs.

Documentation

File	Contents
Aelos Embodied使用手册V1.0.1版-基于镜像1.0.pdf	65-page official manual: aelos_edu panels, serial and Wi-Fi connection, download and run, custom actions, gamepad, sensors, vision, Python editing, TongVerse and image flashing
Aelos_EM视觉使用.pptx	20 slides on the head and chest cameras, RGB vs HSV, the eyedropper HSV picker, colour blocks and window-based visual alignment
Ar_tag码标签使用.pptx	15 slides on AR-tag IDs, ar_track_alvar, X/Y/yaw geometry and alignment/navigation exercises
传感器的使用.pptx	16 slides on LED, touch, photoresistor and temperature/humidity sensors, the ID1/ID2 ports and the generated Python example
ARtag打印文件.pdf	One-page printable AR-tag sheet, IDs 0–8 at 4 cm × 4 cm

All documents are Chinese. They are Leju course material, without a stated open license.

Toolchain and Code Generation

• Robot: Aelos Embodied biped on a D-Robotics RDK X5 controller running Ubuntu with a ROS 1-style catkin_ws and the aelos_smart_ros package. The head camera looks forward and the chest camera looks about 45° down; both return 640×480. A magnetometer, two magnetic sensor ports (ID1/ID2), a speaker and microphone, and the 2.4 GHz gamepad link (channels 1–99) are built in.
• IDE: aelos_edu 2.1.5+ on Windows or macOS with a Chinese UI. A new project must use the robot model Aelos Embodied.
• Deployment: action functions download over USB serial, Python over Wi-Fi; the program then runs from the robot after the serial link is disconnected. Offline Lua download supports neither sensor nor vision blocks.
• Code generation: blocks become Lua for the on-robot runtime (MOTOrigid16, HKEY(), DelayMs) or main.py on the ROS path (base_action, artag_port, sensor_port, nodes). Vision and AR-tag blocks only have working Python generators; their Lua generators are stubs, which is why EM_demo ships a main.py.
• Simulation: TongVerse, Leju's Isaac Sim-based Sim2Real platform (Linux and Docker), is covered in the manual but not included here.

Getting Started

1. Install aelos_edu 2.1.5 or newer from the Leju support page and open a project folder. Keep the .abe, test.xml and user*Action.js files together, because the generated block modules reference each other.
2. Select the robot model Aelos Embodied when creating a project; a mismatched model breaks the serial connection.
3. Download action functions over USB serial and Python over Wi-Fi, disconnect the serial link, then press Run.
4. Import .src routines as custom actions, and copy the matching .mp3 files onto the robot in U-disk mode for music-synchronized performances.
5. Check the serial project first when a program does not start, and confirm the project's IDE version before diagnosing further.

Notes and Limitations

• Project metadata spans IDE versions 2.1.6 (EM_demo), 2.2.1 (EM_default) and 2.2.8 (the scaffolds); open each with a matching aelos_edu version.
• EM_demo/main.lua, lib.lua and Actionlib.lua are Lua 5.3 bytecode consumed by the on-robot runtime. Only the Temp*.lua templates, CustomAction.lua and Musics.lua are readable text.
• EM_serialcheck has no workspace file and try is an empty start block, so both are scaffolds rather than runnable programs.
• .abe metadata records the original authoring path (vendor SD card, engineer desktop, local downloads), so the IDE may ask where to relocate a project.
• No hardware datasheet (degrees of freedom, payload, battery) is included; refer to Leju for the specification.
• The .mp3 assets account for most of the roughly 350 MB checkout.